Saves over 8KB of allocations per new connection.
benchmark old ns/op new ns/op delta
BenchmarkServerFakeConnNoKeepAlive 28777 24927 -13.38%
benchmark old allocs new allocs delta
BenchmarkServerFakeConnNoKeepAlive 52 46 -11.54%
benchmark old bytes new bytes delta
BenchmarkServerFakeConnNoKeepAlive 13716 5286 -61.46%
R=golang-dev, adg
CC=golang-dev
https://golang.org/cl/7799047
After the revert of revision 9ea9e7e6e0c8
the related revision 76ff7da868c6 must be reverted too.
Fixes#5102.
R=golang-dev, r
CC=golang-dev
https://golang.org/cl/7961044
If a handler didn't consume all its Request.Body, child.go was
closing the socket while the host was still writing to it,
causing the child to send a RST and the host (at least nginx)
to send an empty response body.
Now, we tell the host we're done with the request/response
first, and then close our input pipe after consuming a bit of
it. Consuming the body fixes the problem, and flushing to the
host first to tell it that we're done increases the chance
that the host cuts off further data to us, meaning we won't
have much to consume.
No new tests, because this package is lacking in tests.
Tested by hand with nginx. See issue for testing details.
Fixes#4183
R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/7939045
Also adjust the implementation of applyRelocationsAMD64
so that the test added in CL 6848044 still passes.
R=golang-dev, minux.ma
CC=golang-dev
https://golang.org/cl/7686049
XOR key into data 128 bits at a time instead of 64 bits
and pipeline half of state loads. Rotate loop to allow
single-register indexing for state[i].
On a MacBookPro10,2 (Core i5):
benchmark old ns/op new ns/op delta
BenchmarkRC4_128 412 224 -45.63%
BenchmarkRC4_1K 3179 1613 -49.26%
BenchmarkRC4_8K 25223 12545 -50.26%
benchmark old MB/s new MB/s speedup
BenchmarkRC4_128 310.51 570.42 1.84x
BenchmarkRC4_1K 322.09 634.48 1.97x
BenchmarkRC4_8K 320.97 645.32 2.01x
For comparison, on the same machine, openssl 0.9.8r reports
its rc4 speed as somewhat under 350 MB/s for both 1K and 8K
(it is operating 64 bits at a time).
On an Intel Xeon E5520:
benchmark old ns/op new ns/op delta
BenchmarkRC4_128 418 259 -38.04%
BenchmarkRC4_1K 3200 1884 -41.12%
BenchmarkRC4_8K 25173 14529 -42.28%
benchmark old MB/s new MB/s speedup
BenchmarkRC4_128 306.04 492.48 1.61x
BenchmarkRC4_1K 319.93 543.26 1.70x
BenchmarkRC4_8K 321.61 557.20 1.73x
For comparison, on the same machine, openssl 1.0.1
reports its rc4 speed as 587 MB/s for 1K and 601 MB/s for 8K.
R=agl
CC=golang-dev
https://golang.org/cl/7865046
-- amd64 --
On a MacBookPro10,2 (Core i5):
benchmark old ns/op new ns/op delta
BenchmarkRC4_128 470 421 -10.43%
BenchmarkRC4_1K 3123 3275 +4.87%
BenchmarkRC4_8K 26351 25866 -1.84%
benchmark old MB/s new MB/s speedup
BenchmarkRC4_128 272.22 303.40 1.11x
BenchmarkRC4_1K 327.80 312.58 0.95x
BenchmarkRC4_8K 307.24 313.00 1.02x
For comparison, on the same machine, openssl 0.9.8r reports
its rc4 speed as somewhat under 350 MB/s for both 1K and 8K.
The Core i5 performance can be boosted another 20%, but only
by making the Xeon performance significantly slower.
On an Intel Xeon E5520:
benchmark old ns/op new ns/op delta
BenchmarkRC4_128 774 417 -46.12%
BenchmarkRC4_1K 6121 3200 -47.72%
BenchmarkRC4_8K 48394 25151 -48.03%
benchmark old MB/s new MB/s speedup
BenchmarkRC4_128 165.18 306.84 1.86x
BenchmarkRC4_1K 167.28 319.92 1.91x
BenchmarkRC4_8K 167.29 321.89 1.92x
For comparison, on the same machine, openssl 1.0.1
(which uses a different implementation than 0.9.8r)
reports its rc4 speed as 587 MB/s for 1K and 601 MB/s for 8K.
It is using SIMD instructions to do more in parallel.
So there's still some improvement to be had, but even so,
this is almost 2x faster than what it replaced.
-- 386 --
On a MacBookPro10,2 (Core i5):
benchmark old ns/op new ns/op delta
BenchmarkRC4_128 3491 421 -87.94%
BenchmarkRC4_1K 28063 3205 -88.58%
BenchmarkRC4_8K 220392 25228 -88.55%
benchmark old MB/s new MB/s speedup
BenchmarkRC4_128 36.66 303.81 8.29x
BenchmarkRC4_1K 36.49 319.42 8.75x
BenchmarkRC4_8K 36.73 320.90 8.74x
On an Intel Xeon E5520:
benchmark old ns/op new ns/op delta
BenchmarkRC4_128 2268 524 -76.90%
BenchmarkRC4_1K 18161 4137 -77.22%
BenchmarkRC4_8K 142396 32350 -77.28%
benchmark old MB/s new MB/s speedup
BenchmarkRC4_128 56.42 244.13 4.33x
BenchmarkRC4_1K 56.38 247.46 4.39x
BenchmarkRC4_8K 56.86 250.26 4.40x
R=agl
CC=golang-dev
https://golang.org/cl/7547050
Fixes#5061.
Current code relies on the fact that fd's are automatically removed from epoll set when closed. However, it is not true. Underlying file description is removed from epoll set only when *all* fd's referring to it are closed.
There are 2 bad consequences:
1. Kernel delivers notifications on already closed fd's.
2. The following sequence of events leads to error:
- add fd1 to epoll
- dup fd1 = fd2
- close fd1 (not removed from epoll since we've dup'ed the fd)
- dup fd2 = fd1 (get the same fd as fd1)
- add fd1 to epoll = EEXIST
So, if fd can be potentially dup'ed of fork'ed, it's necessary to explicitly remove the fd from epoll set.
R=golang-dev, bradfitz, dave
CC=golang-dev
https://golang.org/cl/7870043
Add missing CLOSUREVAR in switch.
Mark MAKE, string conversion nodes as impossible.
Control statements do not need instrumentation.
Instrument COM and LROT nodes.
Instrument map length.
Update #4228
R=dvyukov, golang-dev
CC=golang-dev
https://golang.org/cl/7504047
Hashtable is arranged as an array of
8-entry buckets with chained overflow.
Each bucket has 8 extra hash bits
per key to provide quick lookup within
a bucket. Table is grown incrementally.
Update #3885
Go time drops from 0.51s to 0.34s.
R=r, rsc, m3b, dave, bradfitz, khr, ugorji, remyoudompheng
CC=golang-dev
https://golang.org/cl/7504044
Found this inconsistency from net/http's Server while
debugging Issue 4183
Unfortunately this package lacks testing around this,
or most of child.go. :/
R=golang-dev, adg
CC=golang-dev
https://golang.org/cl/7735046
No net package involved. And with ReportAllocs we can see
how much garbage is created per request.
R=adg, dave
CC=golang-dev
https://golang.org/cl/7913044
Inserting a key-value pair into a hashmap storing keys or values
indirectly can cause the garbage collector to find the hashmap in
an inconsistent state.
Fixes#5074.
R=golang-dev, minux.ma, rsc
CC=golang-dev
https://golang.org/cl/7913043
- A note doesn't have to be in the first
comment of a comment group anymore, and
several notes may appear in the same comment
group (e.g., it is fairly common to have a
TODO(uid) note immediately following another
comment).
- Define a doc.Note type which also contains
note uid and position info.
- Better formatting in godoc output. The position
information is not yet used, but could be used to
locate the note in the source text if desired.
Fixes#4843.
R=r, cnicolaou
CC=gobot, golang-dev
https://golang.org/cl/7496048
On NetBSD tv_sec is already an int64 so no need for a test.
On OpenBSD, semasleep expects a Unix time as argument,
and 1<<30 is in 2004.
R=golang-dev, iant
CC=golang-dev
https://golang.org/cl/7810044
The example is the same as the FileServer one, but
it's relevant for both.
Also use strings.TrimPrefix while I'm here.
R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/7598046
Now that revision 0c029965805f is in, it's easy
to guarantee that we never access a driver.Conn
concurrently, per the database/sql/driver contract,
so we can remove this overlarge mutex.
Fixes#3857
R=golang-dev, adg
CC=golang-dev
https://golang.org/cl/7707047
If LookPath in Command fails, sets a sticky error, and then
StdinPipe, StdoutPipe, or StderrPipe were called, those pipe
fds were never cleaned up.
Fixes#5071
R=golang-dev, rogpeppe
CC=golang-dev
https://golang.org/cl/7799046
The current SysAlloc implementation suffers from a signed vs unsigned
comparision bug. Since the error code from mmap is negated, the
unsigned comparision of v < 4096 is always false on error. Fix this
by switching to the darwin/freebsd/linux mmap model and leave the mmap
return value unmodified.
R=golang-dev, r
CC=golang-dev
https://golang.org/cl/7870044
Avoids printing it every time we ask a question about the package from
the command line.
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7789048
This provides a way to generate core dumps when people need them.
The settings are:
GOTRACEBACK=0 no traceback on panic, just exit
GOTRACEBACK=1 default - traceback on panic, then exit
GOTRACEBACK=2 traceback including runtime frames on panic, then exit
GOTRACEBACK=crash traceback including runtime frames on panic, then crash
Fixes#3257.
R=golang-dev, devon.odell, r, daniel.morsing, ality
CC=golang-dev
https://golang.org/cl/7666044
CL 7799045 relaxed the restriction in cmd/go on ~ in GOPATH
to allow paths with ~ in the middle while continuing to
protect against the common mistake of using GOPATH='~/home'
instead of GOPATH=~/home. Unfortunately go/build still
filters these paths out:
$ GOPATH=/tmp/test~ing go build
test.go:22:2: cannot find package "test" in any of:
/usr/lib/go/test (from $GOROOT)
($GOPATH not set)
So relax the requirement in go/build, too.
R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/7826043
NEGL does a negation of the bottom 32 bits and then zero-extends to 64 bits,
resulting in a negative 32-bit number but a positive 64-bit number.
NEGQ does a full 64-bit negation, so that the result is negative both as
a 32-bit and as a 64-bit number.
This doesn't matter for the functions that are declared to return int32.
It only matters for the ones that return int64 or void* [sic].
This will fix the current incorrect error in the OpenBSD/amd64 build.
The build will still be broken, but it won't report a bogus error.
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7536046
The database/sql/driver docs make this promise:
"Conn is a connection to a database. It is not used
concurrently by multiple goroutines."
That promises exists as part of database/sql's overall
goal of making drivers relatively easy to write.
So far this promise has been kept without the use of locks by
being careful in the database/sql package, but sometimes too
careful. (cf. golang.org/issue/3857)
The CL associates a Mutex with each driver.Conn, and with the
interface value progeny thereof. (e.g. each driver.Tx,
driver.Stmt, driver.Rows, driver.Result, etc) Then whenever
those interface values are used, the Locker is locked.
This CL should be a no-op (aside from some new Lock/Unlock
pairs) and doesn't attempt to fix Issue 3857 or Issue 4459,
but should make it much easier in a subsequent CL.
Update #3857
R=golang-dev, adg
CC=golang-dev
https://golang.org/cl/7803043
Bring net/fd_linux.go back (it was deleted this morning)
because it is still needed for ARM.
Fix a few typos in the runtime reorg.
R=golang-dev, r
CC=golang-dev
https://golang.org/cl/7759046
thread_GOOS.c becomes os_GOOS.c.
signal_GOOS_GOARCH.c becomes os_GOOS_GOARCH.c,
but with non-GOARCH-specific code moved into os_GOOS.c.
The actual arch-specific signal handler moves into signal_GOARCH.c
to avoid per-GOOS duplication.
New files signal_GOOS_GOARCH.h provide macros for
accessing fields of the very system-specific signal info structs.
Lots moving, but nothing changing.
This is a preliminarly cleanup so I can work on the signal
handling code to fix some open issues without having to
make each change 13 times.
Tested on Linux and OS X, 386 and amd64.
Will fix Plan 9, Windows, and ARM after the fact if necessary.
(Plan 9 and Windows should be fine; ARM will probably have some typos.)
Net effect: -1081 lines of code.
R=golang-dev, r
CC=golang-dev
https://golang.org/cl/7565048
The problem is that new network poller can have spurious
rediness notifications. This implementation ensures that
the socket is actually connected.
R=golang-dev, rsc, akumar
CC=golang-dev
https://golang.org/cl/7785043
The new build tag "go1.1" will be satisfied by any Go 1.z release >= 1.1.
In general, the build tag "go1.x" will be satisfied by any Go 1.z release >= 1.x.
What happens when we reach Go 2 is yet to be decided.
The tags "go1" or "go1.0" are missing, because +build tags did not exist
before then, and also because the Go 1.0 releases do not recognize them.
The new -installsuffix flag gives access to the build context's InstallSuffix
(formerly named InstallTag, but not part of Go 1.0), for use in isolating
builds to custom directories. For example -race implies -installsuffix race,
and an AppEngine-specific build might use -tags appengine -installsuffix appengine.
Fixes#4116.
Fixes#4443.
R=golang-dev, bradfitz, r
CC=golang-dev
https://golang.org/cl/7794043
If a fixed size array is passed in as the decode target and the JSON
to decode has extra array elements that are objects, then previously
the decoder would return a "data changing underfoot" error.
Fixes#3717.
R=golang-dev, adg, rsc
CC=golang-dev
https://golang.org/cl/7490046
Fixes#4705.
Note that libjpeg will print a warning to stderr if there are many
extraneous bytes, but can be silent if the extraneous bytes can fit
into its int32 bit-buffer for Huffman decoding. I'm guessing that
this is why whatever encoder that produced the image filed for issue
4705 did not realize that they are, strictly speaking, generating an
invalid JPEG. That issue's attached image has two extraneous bytes.
For example, piping the program below into libjpeg's djpeg program
will print an "18 extraneous bytes" warning, even though N == 20.
$ cat main.go
package main
import (
"bytes"
"image"
"image/color"
"image/jpeg"
"os"
)
const N = 20
func main() {
// Encode a 1x1 red image.
m := image.NewRGBA(image.Rect(0, 0, 1, 1))
m.Set(0, 0, color.RGBA{255, 0, 0, 255})
buf := new(bytes.Buffer)
jpeg.Encode(buf, m, nil)
b := buf.Bytes()
// Strip the final "\xff\xd9" EOI marker.
b = b[:len(b)-2]
// Append N dummy 0x80 bytes to the SOS data.
for i := 0; i < N; i++ {
b = append(b, 0x80)
}
// Put back the "\xff\xd9" EOI marker.
b = append(b, 0xff, 0xd9)
os.Stdout.Write(b)
}
$ go run main.go | djpeg /dev/stdin > /tmp/foo.pnm
Corrupt JPEG data: 18 extraneous bytes before marker 0xd9
The resultant /tmp/foo.pnm is a perfectly good 1x1 red image.
R=r
CC=golang-dev
https://golang.org/cl/7750043
With the global redefinition of runtime·open by CL 7543043,
we need to provide a third argument and remove the cast
to the string.
Fixes build on 386 version of Plan 9.
R=khr, rsc, rminnich, ality
CC=golang-dev
https://golang.org/cl/7644047
The parameter list layout function was incorrectly computing the
end of the previous line in cases where a parameter type spanned
multiple lines. As a result, an extra (valid, but not needed)
comma was introduced before the paremeter list's closing parenthesis.
Fixes#4533.
R=rsc
CC=golang-dev
https://golang.org/cl/7674044
This was a bug that didn't manifest itself before CL 7674044;
but with that CL and without this fix, the go/doc tests fail.
(The bug fixed by 7674044 and the bug fixed here cancelled
each other out w/ respect to the go/doc tests).
R=rsc
CC=golang-dev
https://golang.org/cl/7628045
Uses AES hardware instructions on 386/amd64 to implement
a fast hash function. Incorporates a random key to
thwart hash collision DOS attacks.
Depends on CL#7548043 for new assembly instructions.
Update #3885
Helps some by making hashing faster. Go time drops from
0.65s to 0.51s.
R=rsc, r, bradfitz, remyoudompheng, khr, dsymonds, minux.ma, elias.naur
CC=golang-dev
https://golang.org/cl/7543043
If two fields have the same name but different explicit name spaces,
treat as non-conflicting. This allows parsing common XML formats
that have ns1:tag and ns2:tag in the same XML element.
Fixes#4691.
Allow setting the default name space for unadorned tags, by
writing to Decoder.DefaultSpace. This allows turned the job of
parsing common XML formats that have tag and ns2:tag in the
same XML element into the first case by setting DefaultSpace="ns1".
Fixes#3703.
Use name space attributes when decoding.
Attach name space to attributes when encoding.
Could be done with fewer annotations, but semantically correct as is.
Fixes#3526.
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7227056
The issue was that scvg is assigned *after* the scavenger goroutine is started,
so when the scavenger calls entersyscall() the g==scvg check can fail.
Fixes#5025.
R=golang-dev, iant
CC=golang-dev
https://golang.org/cl/7629045
The problem is that there are lots of dead G's from previous tests,
each dead G consumes 1 stack segment.
Fixes#5034.
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7749043
Adds a new reader to filter newlines, which fixes errors seen in the
decoder chunking code. Found additional issues with whitespace handling
after the first padding character.
Fixes#4779.
R=minux.ma, rsc, bradfitz
CC=golang-dev
https://golang.org/cl/7311069
The system call takes an int, but the kernel stores it in a uint16.
At least one Linux system sets /proc/sys/net/core/somaxconn
to 262144, which ends up being 0 in the uint16. Avoid being tricked.
FreeBSD sources also store the backlog in a uint16.
Assume the problem is systemic and fix it everywhere.
Fixes#5030.
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7480046
The old code just assumed that the only thing
you can embed is a struct. Not true.
Fixes#3803.
R=golang-dev, adg
CC=golang-dev
https://golang.org/cl/7743043
This isn't as bad as it used to be, but add a bit
more detail to close the issue.
Fixes#3359
R=golang-dev, adg
CC=golang-dev
https://golang.org/cl/7606044
Edit encoding/base64's internals and tests to match encoding/base32.
Properly handling line breaks in padding is left for another CL.
R=dsymonds
CC=golang-dev
https://golang.org/cl/7693044
For expressions where the result type is independent
of the argument types (comparisons, conversions, rhs
of shifts), set the final expression types for those
subtrees early.
This fixes several bugs where incorrect lhs shift
operands where used (say in a comparison), but were
not reported.
Together with the changes listed below this CL fixes
many type-checker bugs.
Also:
- better documented updateExprType
- added larger comment to expr.go explaining
the basic expression checking algorithm
- use latest definition for indices and make
arguments; use the same code to check both
- use the same mechanism for cycle detection
in constant expressions as for variables
(new field Constant.visited)
- more tests for complex and make builtins
- many more and systematic tests for shifts;
moved them into separate testfile
- in the testing code, don't compare the
expected error pattern against itself
(the actual message was always ignored...)
- fix affected error patterns in the test files
- various cleanups along the way
R=adonovan
CC=golang-dev
https://golang.org/cl/7432051
This includes a simplified resolvePath function and tests for all normal and abnormal path resolution examples described in RFC 3986, sections 5.4.1 and 5.4.2 [1]. Some of those examples failed before (see http://play.golang.org/p/F0ApSaXniv).
Also, parsing a reference "//foo" now works as expected. It was treated as an absolute path with very weird results (see http://play.golang.org/p/089b-_xoNe).
During path resolution, all dot segments are removed as described by the RFC.
A few existing tests had to be changed because they expected the wrong output.
Fixes#4700.
Fixes#4706.
[1] http://tools.ietf.org/html/rfc3986#section-5.4.1
R=rsc, adg, bradfitz
CC=golang-dev
https://golang.org/cl/7203059
Previously it was evaluated once, so re-using the timeout option
repeatedly would always generate the same deadine.
Also switch to doing just one pass over the options, making
the private interface actually useful.
R=golang-dev, dsymonds
CC=golang-dev
https://golang.org/cl/7608045
Update #4929
Regenerated from FreeBSD-9.1 for amd64 and 386, FreeBSD-CURRENT for arm.
R=devon.odell, minux.ma, rsc
CC=golang-dev
https://golang.org/cl/7438053
The call to the C function runtime.findnull() requires
that we provide the argument at 0(SP).
R=rsc, rminnich, ality
CC=golang-dev
https://golang.org/cl/7559047
For better printing, I recently changed Name to return "<unnamed>" for templates
with empty names, but this causes trouble for the many packages that used "" as
the template name, so restore the old behavior.
It's usually printed as a quoted string anyway, so it should be fine.
R=golang-dev, dsymonds
CC=golang-dev
https://golang.org/cl/7577044
Otherwise the next goroutine run on the m
can get inadvertently locked if it executes a cgo call
that turns on the internal lock.
While we're here, fix the cgo panic unwind to
decrement m->ncgo like the non-panic unwind does.
Fixes#4971.
R=golang-dev, iant, dvyukov
CC=golang-dev
https://golang.org/cl/7627043
Now the default startup is that the program begins at _rt0_386_$GOOS,
which behaves as if calling main(argc, argv). Main jumps to _rt0_386.
This makes the _rt0_386 entry match the expected semantics for
the standard C "main" function, which we can now provide for use when
linking against a standard C library.
386 analogue of https://golang.org/cl/7525043
R=golang-dev, r
CC=golang-dev
https://golang.org/cl/7551045
Implement what Chrome calls socket "late binding". See:
https://insouciant.org/tech/connection-management-in-chromium/
In a nutshell, if our HTTP client needs a TCP connection to a
remote host and there's not an idle one available, rather than
kick off a dial and wait for that specific dial, we instead
kick off a dial and wait for either our own dial to finish, or
any other TCP connection to that same host to become
available.
The implementation looks like a classic "Learning Go
Concurrency" slide.
Chrome's commit and numbers:
http://src.chromium.org/viewvc/chrome?view=rev&revision=36230
R=golang-dev, daniel.morsing, adg
CC=golang-dev
https://golang.org/cl/7587043
Change 231af8ac63aa (CL 7314062) made runtime.enteryscall()
set m->mcache = nil, which means that we can no longer use
syscall.errstr in syscall.Syscall and syscall.Syscall6, since it
requires a new buffer to be allocated for holding the error string.
Instead, we use pre-allocated per-M storage to hold error strings
from syscalls made while in entersyscall mode, and call
runtime.findnull to calculate the lengths.
Fixes#4994.
R=rsc, rminnich, ality, dvyukov, rminnich, r
CC=golang-dev
https://golang.org/cl/7567043
This CL splits multicast listener tests into two; for IPv4 and
for IPv6. It also removes redundant test inputs and makes sure
that assignment of multicast interface to stablize the tests.
Fixes#4059.
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7565043
Move pollServer from fd_unix.go to fd_poll_unix.go.
Add pollServerInit(*NetFD) to allow custom initialization.
Add pollServer.Close(*NetFD) to allow custom finalization.
Move setDeadline() to fd_poll_unix.go to allow custom handling of deadlines.
Move newPollServer() to fd_poll_unix.go to allow custom initialization.
No logical code changes.
The next step will be to turn off fd_poll_unix.go for some platform
(I have changes for darwin/linux) and redirect it into runtime. See:
https://golang.org/cl/7569043/diff/2001/src/pkg/net/fd_poll_runtime.go
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7513045
The deadlock episodically occurs on misc/cgo/test/TestCthread.
The problem is that starttheworld() leaves some P's with local work
without M's. Then all active M's enter into syscalls, but reject to
wake another M's due to the following check (both in entersyscallblock() and in retake()):
if(p->runqhead == p->runqtail &&
runtime·atomicload(&runtime·sched.nmspinning) +
runtime·atomicload(&runtime·sched.npidle) > 0)
continue;
R=rsc
CC=golang-dev
https://golang.org/cl/7424054
Still to do: non-linux and non-amd64.
It may work on other ELF-based amd64 systems too, but untested.
"go test -ldflags -hostobj $GOROOT/misc/cgo/test" passes.
Much may yet change, but this seems a reasonable checkpoint.
R=iant
CC=golang-dev
https://golang.org/cl/7369057
Ensure that accept/connect respect deadline,
even if the operation can be executed w/o blocking.
Note this changes external behavior, but it makes
it consistent with read/write.
Factor out deadline check into pollServer.PrepareRead/Write,
in preparation for edge triggered pollServer.
Ensure that pollServer.WaitRead/Write are not called concurrently
by adding rio/wio locks around connect/accept.
R=golang-dev, mikioh.mikioh, bradfitz, iant
CC=golang-dev
https://golang.org/cl/7436048
Some IDS somewhere thinks "Go http package" is a virus.
Make it something else for Go 1.1. Dumb but easy.
R=golang-dev, bradfitz, minux.ma
CC=golang-dev
https://golang.org/cl/7532043
Now the default startup is that the program begins at _rt0_amd64_$GOOS,
which sets DI = argc, SI = argv and jumps to _rt0_amd64.
This makes the _rt0_amd64 entry match the expected semantics for
the standard C "main" function, which we can now provide for use when
linking against a standard C library.
R=golang-dev, devon.odell, minux.ma
CC=golang-dev
https://golang.org/cl/7525043
Previously the HTTP client's (*Response).Body.Close would try
to keep reading until EOF, hoping to reuse the keep-alive HTTP
connection, but the EOF might never come, or it might take a
long time. Now we immediately close the TCP connection if we
haven't seen EOF.
This shifts the burden onto clients to read their whole response
bodies if they want the advantage of reusing TCP connections.
In the future maybe we could decide on heuristics to read some
number of bytes for some max amount of time before forcefully
closing, but I'd rather not for now.
Statistically, touching this code makes things regress, so I
wouldn't be surprised if this introduces new bugs, but all the
tests pass, and I think the code is simpler now too. Maybe.
Please test your HTTP client code before Go 1.1.
Fixes#3672
R=golang-dev, adg
CC=golang-dev
https://golang.org/cl/7419050
broke arm garbage collector
traceback_arm fails with a missing pc. It needs CL 7494043.
But that only makes the build break later, this time with
"invalid freelist". Roll back until it can be fixed correctly.
««« original CL description
runtime: restrict stack root scan to locals and arguments
R=rsc
CC=golang-dev
https://golang.org/cl/7301062
»»»
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7493044
On OpenBSD 5.2, calling getsockname on an unbound Unix domain socket
results in a successful syscall, however the AF is unset and the length
is returned as zero. This has been changed to more portable behaviour,
which will be included in the OpenBSD 5.3 release.
For now, work around this by treating a successful getsockname() call
that returns a family of AF_UNSPEC and length of zero as a AF_UNIX
socket.
Makes TestPassFD work on OpenBSD 5.2.
Fixes#4956.
R=golang-dev, minux.ma, rsc, mikioh.mikioh
CC=golang-dev
https://golang.org/cl/7449046
Implementation closely based on Russ' CL 7440047.
Future work: The error messages could be better
(e.g., instead of "missing return" it might say
"missing return (no default in switch)", etc.).
R=adonovan, rsc
CC=golang-dev
https://golang.org/cl/7437049
Use a 17 MB payload instead of a 1 MB payload, since
OS X can apparently buffer up to 16 MB in its pipes.
Fixes#4958 maybe
R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/7453049
If the constant CollectStats is non-zero and GOGCTRACE=1
the garbage collector will print basic statistics about executed
GC instructions.
R=golang-dev, dvyukov
CC=golang-dev, rsc
https://golang.org/cl/7413049
The old code would decode the palette only for 8-bit
images during a DecodeConfig.
This CL keeps the behavior for 8-bit images and sets
up the decoded palette also for 1, 2 and 4-bit images.
Fixes#4279.
R=golang-dev, nigeltao
CC=golang-dev
https://golang.org/cl/7421048
Fixes#4904.
The problem was that when the test runs the heap had grown to ~100MB,
so GC allows it to grow to 200MB, and so the test fails.
Moving the test to a separate process makes it much more isolated and stable.
R=golang-dev, minux.ma
CC=golang-dev
https://golang.org/cl/7441046
Also make the crossover point an architecture-dependent constant,
although it's the same everywhere for now.
BenchmarkAppendStr1Byte 416 145 -65.14%
BenchmarkAppendStr4Bytes 743 217 -70.79%
BenchmarkAppendStr8Bytes 421 270 -35.87%
BenchmarkAppendStr16Bytes 415 403 -2.89%
BenchmarkAppendStr32Bytes 415 391 -5.78%
R=golang-dev, iant
CC=golang-dev
https://golang.org/cl/7459044
Bridge methods for embedded interfaces were
passing the interface twice: once as receiver,
once as first param.
Covered by $GOROOT/test/ddd.go.
Also:
- invent names ("arg%d") for parameters if missing.
- refactoring: move common code for bridge methods into
createParams and emitTailCall.
R=gri
CC=golang-dev
https://golang.org/cl/7437047
Putting the M initialization in multiple places will not scale.
Various code assumes mstart is the start already. Make it so.
R=golang-dev, devon.odell
CC=golang-dev
https://golang.org/cl/7420048
There is a #pragma dynexport crosscall2, to help SWIG,
and 6l cannot export the symbol if it doesn't get to see it.
R=golang-dev, iant
CC=golang-dev
https://golang.org/cl/7448044
Now that untyped expressions are done in two phases, the
identity of operand.expr is used as a map key; when reusing
operand values we now must be careful to update the
expr field.
R=gri
CC=golang-dev
https://golang.org/cl/7444049
net/http/cookiejar.
This is a straight rename. There are no code changes.
Fixes#1960.
R=rsc, adg
CC=dr.volker.dobler, golang-dev
https://golang.org/cl/7436043
Completely rethought shift expression type checking.
Instead of attempting to type-check them eagerly, now
delay the checking of untyped constant lhs in non-
constant shifts until the final expression type
becomes clear. Once it is clear, update the respective
expression tree with the final (not untyped) type and
check respective shift lhs' where necessary.
This also cleans up another conundrum: How to report
the type of untyped constants as it changes from
untyped to typed. Now, Context.Expr is only called
for an expresion x once x has received its final
(not untyped) type (for constant initializers, the
final type may still be untyped).
With this CL all remaining std lib packages that
did not typecheck due to shift errors pass now.
TODO: There's a lot of residual stuff that needs
to be cleaned up but with this CL all tests pass
now.
R=adonovan, axwalk
CC=golang-dev
https://golang.org/cl/7381052
The Name field of the stat structure is variable length
and the marshalling code in package syscall requires
a buf long enough to contain the Name as well as the
static data. This change makes sure that the buffer in
os.Rename is allocated with the appropriate length.
R=rsc, rminnich, ality, r
CC=golang-dev
https://golang.org/cl/7453044
There are some function pointers declared by 6c in
package runtime without initialization and then also
declared in package runtime/cgo with initialization,
so that if runtime/cgo is linked in, the function pointers
are non-nil, and otherwise they are nil. We depend on
this property for implementing non-essential cgo hooks
in package runtime.
The declarations in package runtime are 6c-compiled
and end up in .6 files. The declarations in package runtime/cgo
are gcc-compiled and end up in .o files. Since 6l links the .6
and .o files together, this all works.
However, when we switch to "external linking" mode,
6l will not see the .o files, and it would be up to the host linker
to resolve the two into a single initialized symbol.
Not all host linkers will do this (in particular OS X gcc will not).
To fix this, move the cgo declarations into 6c-compiled code,
so that they end up in .6 files, so that 6l gets them no matter what.
R=golang-dev
CC=golang-dev
https://golang.org/cl/7440045
The naming in this package is a disaster.
Make it all consistent.
Remove some 'static' from functions that will
be referred to from other files soon.
This CL is purely renames using global search and replace.
Submitting separately so that real changes will not
be drowned out by these renames in future CLs.
TBR=iant
CC=golang-dev
https://golang.org/cl/7416046
runtime: double-check that symbol table is sorted
If the symbol table is unsorted, the binary search in findfunc
will not find its func, which will make stack traces stop early.
When the garbage collector starts using the stack tracer,
that would be a serious problem.
The unsorted symbol addresses came from from two things:
1. The symbols in an ELF object are not necessarily sorted,
so sort them before adding them to the symbol list.
2. The __i686.get_pc_thunk.bx symbol is present in multiple
object files and was having its address adjusted multiple
times, producing an incorrect address in the symbol table.
R=golang-dev, iant
CC=golang-dev
https://golang.org/cl/7440044
This reveals the presence of control and non-printable characters in the
errors returned by the Parse functions. Also add unit tests for NumError.
R=golang-dev, r, rsc
CC=golang-dev
https://golang.org/cl/7393075
The test suite for ReadResponse was not checking the error return on the io.Copy
on the body. This was masking two errors: the handling of chunked responses to
HEAD requests and the handling of Content-Length > 0 to HEAD.
The former manifested itself as an 'unexpected EOF' when doing the io.Copy
because a chunked reader was assigned but there were no chunks to read. The
latter cause (*http.Response).Write to report an error on HEAD requests
because it saw a Content-Length > 0 and expected a body.
There was also a missing \r\n in one chunked test that meant that the chunked
encoding was malformed. This does not appear to have been intentional.
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7407046
Revision 6a88e1893941 corrupts the argument to
racefuncenter by pushing the data block pointer
to the stack.
Fixes#4885.
R=dvyukov, rsc
CC=golang-dev
https://golang.org/cl/7381053
syscall: Use NewError for all system errors and introduce
some new errors for compatibility with other packages
and proper error handling in net. Also introduce
Temporary and Timeout methods on ErrorString.
net: Make errors from dial, accept, listen functions follow the
OpError standard and discern whether the underlying
error came from syscall. Since Plan 9 uses a correspondence
between file and network operations, all system error
reporting happens through the underlying file operation.
In Go code, we go through package os for file operations,
so there is another level of indirection in error types.
This change allows us to compare the errors with those in
package syscall, when appropriate.
os: Just use the error string already present in package os,
instead of calling out to package syscall.
R=rsc, ality, rminnich, bradfitz
CC=golang-dev
https://golang.org/cl/7398054
Map order is non-deterministic. Introduce a new
environment string array that tracks the env map.
This allows us to produce identical results for
Environ() upon successive calls, as expected by the
TestConsistentEnviron test in package os.
R=rsc, ality, rminnich, bradfitz, r
CC=golang-dev
https://golang.org/cl/7411047
New tests added for port handling and IDNA domains.
A new test case contains several redundant
tests but provides a nice documentation of the
implemented rules for domain handling.
R=nigeltao
CC=golang-dev
https://golang.org/cl/7393070
Permits all sorts of custom HTTP timeout policies without
adding a new Transport timeout Duration for each combination
of HTTP phases.
This keeps track internally of which TCP connection a given
Request is on, and lets callers forcefully close the TCP
connection for a given request, without actually getting
the net.Conn directly.
Additionally, a future CL will implement res.Body.Close (Issue
3672) in terms of this.
Update #3362
Update #3672
R=golang-dev, rsc, adg
CC=golang-dev
https://golang.org/cl/7372054
ssadump:
- permit naming a package (not just *.go files) on command line.
- set BuildSerially flag when setting Log* flags
(Q. should instead the logging functions take a lock?)
Builder:
- fixed bug when calling variadic function with zero '...'-params.
Added regression test.
interp:
- more external functions:
the 'error' interface
bytes.{Equal,IndexByte}
reflect.(Value).{Bool,NumOut,Out}
syscall.{Close,Fstat,Read,Open,Stat,Lstat,Fstat,
Getdents,ParseDirents,Getwd}
- permit comparisons between *Function and *closure.
With this CL, ssadump can now interpret ssadump itself (!),
loading, parsing, typing, SSA-building, and running
println("Hello, World!"). While a fmt-based equivalent still
lacks some external routines, e.g. math/big, I think there are
diminishing returns in expanding the interpreter (and
debugging it is starting to feel like "Inception").
I'm pretty confident this package is now good enough for wider use.
R=gri
CC=golang-dev
https://golang.org/cl/7392053
add per-P cache of dead G's
add global runnable queue (not used for now)
add list of idle P's (not used for now)
R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/7397061
While submitting CL 7371051 I accidentally reverted much of CL
7395052. This change restores it.
R=gri
TBR=gri
CC=golang-dev
https://golang.org/cl/7364051
The correct semantics of named result parameters and deferred
procedures cannot be implemented with the existing Ret
instruction alone, since the required sequence is:
(1) evaluate return operands and parallel-assign them to
named result parameters
(2) invoke deferred procedures
(3) load named result parameters to form result tuple.
We introduce a new 'rundefers' instruction that explicitly
invokes the deferred procedure calls, and we generate code
that follows the sequence above.
Most functions do not use deferred procedures but this cannot
be known in a single pass. So, we add an optimisation to
eliminate redundant 'rundefers'; it is piggybacked on the
existing pass done for "lifting".
Added tests.
R=gri
CC=golang-dev
https://golang.org/cl/7411043
The current code uses 64-bit pc-relative on 64-bit systems,
but in ELF linkers there is no such thing, so we cannot
express this in a .o file. Change to 32-bit.
R=golang-dev, iant
CC=golang-dev
https://golang.org/cl/7383055
This CL changes the encoding used for the Go symbol table,
stored in the binary and used at run time. It does not change
any of the semantics or structure: the bits are just packed
a little differently.
The comment at the top of runtime/symtab.c describes the new format.
Compared to the Go 1.0 format, the main changes are:
* Store symbol addresses as full-pointer-sized host-endian values.
(For 6g, this means addresses are 64-bit little-endian.)
* Store other values (frame sizes and so on) varint-encoded.
The second change more than compensates for the first:
for the godoc binary on OS X/amd64, the new symbol table
is 8% smaller than the old symbol table (1,425,668 down from 1,546,276).
This is a required step for allowing the host linker (gcc) to write
the final Go binary, since it will have to fill in the symbol address slots
(so the slots must be host-endian) and on 64-bit systems it may
choose addresses above 4 GB.
R=golang-dev, iant
CC=golang-dev
https://golang.org/cl/7403054
Actually it already worked since the spec only requires that
the one immediately preceding a for/switch/... be usable as
the target of a break or continue statement.
Added a test.
Also: allocate Function.lblocks on first use.
R=gri
CC=golang-dev
https://golang.org/cl/7365058
And fix test. Pointer to unsafe.Pointer tests nothing important...
Also identify the incorrect type: go/types.Type is a Stringer.
Also fix a couple of incorrect format verbs found by new printf checker,
now that we can run it on more files.
R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/7385051
The previous approach desugared the ast.SelectorExpr
to make implicit field selections explicit. But:
1) it was clunky since it required allocating temporary
syntax trees.
2) it was not thread-safe since it required poking
types into the shared type map for the new ASTs.
3) the desugared syntax had no place to represent the
package lexically enclosing each implicit field
selection, so it was as if they all occurred in the
same package as the explicit field selection.
This meant unexported field names changed meaning.
This CL does what I should have done all along: just
generate the SSA instructions directly from the original
AST and the promoted field information.
Also:
- add logStack util for paired start/end log messages.
Useful for debugging crashes.
R=gri
CC=golang-dev
https://golang.org/cl/7395052
Adjust the exit status string for Plan 9.
Upon allocating >100 file descriptors, Plan 9
raises a warning. Moreover, the Go runtime for
32-bit version of Plan 9 keeps /dev/bintime
open for its implementation of runtime.nanotime().
This change accounts for these things in
TestExtraFiles.
R=rsc, rminnich, ality, bradfitz
CC=golang-dev
https://golang.org/cl/7363056
Context.Alignof/Offsetsof/Sizeof now provide means
to customize the type checker for a given platform.
- provide Context.Offsetsof to specify the
offsets of struct fields
- use the correct sizes for ints, uint, uintptrs
in constant computations
- moved all size computations into separate file
(sizes.go)
- fixed a bug with string constant slicing
R=adonovan, axwalk
CC=golang-dev
https://golang.org/cl/7363054
- imported objects don't have position information
- gc exported data contains non-exported objects at
the top-level, guard against them
- better error message when dot-imports conflict
with local declarations
R=adonovan, r
CC=golang-dev
https://golang.org/cl/7379052
pkg/go/doc: move BUG notes from Package.Bugs to the general Package.Notes field.
Removing .Bugs would break existing code so it's left in for now.
R=gri, gri, gary.burd, dsymonds, rsc, kevlar
CC=golang-dev
https://golang.org/cl/7341053
Separates the implementation of nanotime on 64-bit
version of Plan 9 from that on the 32-bit version.
The former uses a syscall.
R=rsc, rminnich, ality
CC=golang-dev
https://golang.org/cl/7379051
Functions for representing network connections as files
and vice versa, on Plan 9.
Representing network connections as files is not so
straight-forward, because a network connection on Plan 9
is represented by a host of files rather than a single
file descriptor (as is the case on UNIX). We use the
type system to distinguish between listeners and
connections, returning the control file in the former
case and the data file in the latter case.
R=rsc, rminnich, ality, akumar, bradfitz
CC=golang-dev
https://golang.org/cl/7235068
This avoids an allocation when closures are used
as "macros", in Walk idioms, or as argument to defer.
benchmark old ns/op new ns/op delta
BenchmarkSearchWrappers 1171 354 -69.77%
BenchmarkCallClosure 3 3 -12.54%
BenchmarkCallClosure1 119 7 -93.95%
BenchmarkCallClosure2 183 74 -59.18%
BenchmarkCallClosure3 187 75 -59.57%
BenchmarkCallClosure4 187 76 -58.98%
Compared to Go 1:
benchmark old ns/op new ns/op delta
BenchmarkSearchWrappers 3208 354 -88.97%
Fixes#3520.
R=daniel.morsing, bradfitz, minux.ma, dave, rsc
CC=golang-dev
https://golang.org/cl/7397056
Syscalls return `-1' on error and the representation is always
32-bits. The `$-1' literal in 64-bit assembly is always the
64-bit representation. So this change makes sure that we
always do a 32-bit comparison when checking for error.
Also makes sure that in the error case, we return a 64-bit
`-1' from runtime.seek.
Fixes the arithmetic for handling the error-string in
runtime.Syscall6.
R=golang-dev, rminnich, rsc, ality, minux.ma
CC=golang-dev
https://golang.org/cl/7399052
sigprocmask() is process-wide on darwin, so two concurrent
libcgo_sys_thread_start() can result in all signals permanently
blocked, which in particular blocks handling of nil derefs.
Fixes#4833.
R=golang-dev, dave, rsc
CC=golang-dev
https://golang.org/cl/7324058
It's accessed without mutex protection
in a different goroutine from the one that
sets it.
Also make sure that Client.Call after Client.Close
will reliably return ErrShutdown, and that clients
see ErrShutdown rather than io.EOF when appropriate.
Suggestions welcome for a way to reliably test
the mutex issue.
R=r, iant
CC=golang-dev
https://golang.org/cl/7338045
The heuristics for BitLen of a product of randomly generated primes
are wrong, and the generated candidates never match the required
size for nprimes > 10. This corner case is not expected to be used in
practice.
R=agl
CC=golang-dev
https://golang.org/cl/7397052
This CL allows to receive network interface arrival and depature
notifications through routing sockets on BSD variants. So far
Darwin doesn't support this feature.
Also does small simplification.
Update #4866.
R=golang-dev, lucio.dere, dave
CC=golang-dev
https://golang.org/cl/7365055
Plan 9 I/O preserves message boundaries, while Go
library code is written for UNIX-like operating
systems which do not. Avoid doing zero-length
writes in package os.
R=rsc, rminnich, ality, rminnich, r
CC=golang-dev
https://golang.org/cl/7406046
For now, all the callbacks from C use top-level Go functions,
so they use the equivalent C function pointer, and will continue
to do so. But perhaps some day this will be useful for calling
a Go func value (at least if the type is already known).
More importantly, the Windows callback code needs to be able
to use cgocallback_gofunc to call a Go func value.
Should fix the Windows build.
R=ken2
CC=golang-dev
https://golang.org/cl/7388049
We wrap the final '...' argument's type in types.Slice.
Added tests.
Also:
- Function.writeSignature: suppress slice '[]' when printing
variadic arg '...'.
- Eliminate Package.ImportPath field; redundant
w.r.t. Package.Types.Path.
- Use "TODO: (opt|fix)" notation more widely.
- Eliminate many redundant/stale TODOs.
R=gri
CC=golang-dev
https://golang.org/cl/7378057
Change ARM context register to R7, to get out of the way
of the register allocator during the compilation of the
prologue statements (it wants to use R0 as a temporary).
Step 2 of http://golang.org/s/go11func.
R=ken2
CC=golang-dev
https://golang.org/cl/7369048
This makes Go's quoted-printable decoder more like other
popular ones, allowing through a bare \r or \n, and also
passes through \r\n which looked like a real bug before.
Fixes#4771
R=minux.ma
CC=golang-dev
https://golang.org/cl/7300092
By appending to the wrong (always empty) list, only the last
anonymous field was being considered for promotion.
Also:
- eliminated "function-local NamedTypes" TODO; nothing to do.
- fixed Function.DumpTo: printing of anon receivers was "( T)",
now "(T)"; extracted writeSignature into own function.
- eliminated blockNames function;
thanks to BasicBlock.String, "%s" of []*BasicBlock is fine.
- extracted buildReferrers into own function.
exp/ssa can now build its own transitive closure.
R=gri
CC=golang-dev
https://golang.org/cl/7384054
runtime: add context argument to gogocall
Too many other things use AX, and at least one
(stack zeroing) cannot be moved onto a different
register. Use the less special DX instead.
Preparation for step 2 of http://golang.org/s/go11func.
Nothing interesting here, just split out so that we can
see it's correct before moving on.
R=ken2
CC=golang-dev
https://golang.org/cl/7395050
- append: nothing to do (nonsemantic change).
- delete: now performs correct conversion (+ test).
- emitCompare: nothing to do.
- emitArith (shifts): nothing to do (+ test).
- "banish untyped types": give up on that.
- real, imag: now do correct conversions.
- added comment to interp.go re zero-size values.
R=gri
CC=golang-dev
https://golang.org/cl/7391046
Revision 5e7fd762f356 has changed the output formatting in a way that
is no longer in line with the format described by the revision
ff0ade0b937b which has introduced this functionality.
When decorating the first line, instead of indenting the whole line,
the current implementation adds indentation right after the "decorate"
part and before the "log" message.
The fix addresses this issue.
R=golang-dev, iant, minux.ma, r, rsc, remyoudompheng
CC=golang-dev
https://golang.org/cl/7304094
Previously, the func structure contained an inaccurate value for
the args member and a 0 value for the locals member.
This change populates the func structure with args and locals
values computed by the compiler. The number of args was
already available in the ATEXT instruction. The number of
locals is now passed through in the new ALOCALS instruction.
This change also switches the unit of args and locals to be
bytes, just like the frame member, instead of 32-bit words.
R=golang-dev, bradfitz, cshapiro, dave, rsc
CC=golang-dev
https://golang.org/cl/7399045
The interpreter's os.Exit now triggers a special panic rather
than kill the test process. (It's semantically dubious, since
it will run deferred routines.) Interpret now returns its
exit code rather than calling os.Exit.
Also:
- disabled parts of a few $GOROOT/tests via os.Getenv("GOSSAINTERP").
- remove unnecessary 'slots' param to external functions; they
are never closures.
Most of the tests are disabled until go/types supports shifts.
They can be reenabled if you patch this workaround:
https://golang.org/cl/7312068
R=iant, bradfitz
CC=golang-dev, gri
https://golang.org/cl/7313062
By avoiding the need for self-loops following calls to panic,
we reduce the number of basic blocks considerably.
R=gri
CC=golang-dev, iant
https://golang.org/cl/7403043
Overview: Function.finish() now invokes the "lifting" pass which replaces local allocs and loads and stores to such cells by SSA registers. We use very standard machinery:
(1) we build the dominator tree for the function's control flow graph (CFG) using the "Simple" Lengauer-Tarjan algorithm. (Very "simple" in fact: even simple path compression is not yet implemented.)
In sanity-checking mode, we cross check the dominator tree against an alternative implementation using a simple iterative dataflow algorithm.
This all lives in dom.go, along with some diagnostic printing routines.
(2) we build the dominance frontier for the entire CFG using the Cytron et al algorithm. The DF is represented as a slice of slices, keyed by block index. See buildDomFrontier() in lift.go.
(3) we determine for each Alloc whether it can be lifted: is it only subject to loads and stores? If so, we traverse the iterated dominance frontier (IDF) creating φ-nodes; they are not prepended to the blocks yet.
See liftAlloc() in lift.go.
(4) we perform the SSA renaming algorithm from Cytron et al, replacing all loads to lifted Alloc cells by the value stored by the dominating store operation, and deleting the stores and allocs. See rename() in lift.go.
(5) we eliminate unneeded φ-nodes, then concatenate the remaining ones with the non-deleted instructions of the block into a new slice. We eliminate any lifted allocs from Function.Locals.
To ease reviewing, I have avoided almost all optimisations at this point, though there are many opportunities to explore. These will be easier to understand as follow-up changes.
All the existing tests (pending CL 7313062) pass. (Faster!)
Details:
"NaiveForm" BuilderMode flag suppresses all the new logic.
Exposed as 'ssadump -build=N'.
BasicBlock:
- add .Index field (b.Func[b.Index]==b), simplifying
algorithms such as Kildall-style dataflow with bitvectors.
- rename the Name field to Comment to better reflect its
reduced purpose. It now has a String() method.
- 'dom' field holds dominator tree node; private for now.
- new predIndex method.
- hasPhi is now a method
dom.go:
- domTree: a new struct for a node in a dominator tree.
- buildDomTree builds the dominator tree using the simple
variant Lengauer/Tarjan algorithm with Georgiadis'
bucket optimizations.
- sanityCheckDomTree builds dominance relation using
Kildall-style dataflow and ensures the same result is
obtained.
- printDomTreeDot prints the CFG/DomTree in GraphViz format.
blockopt.go:
- perform a mark/sweep pass to eliminate unreachable
cycles; the previous prune() opt would only eliminate
trivially dead blocks. (Needed for LT algo.)
- using .Index, fuseblocks can now delete fused blocks directly.
- delete prune().
sanity.go: more consistency checks:
- Phi with missing edge value
- local Alloc instructions must appear in Function.Locals.
- BasicBlock.Index, Func consistency
- CFG edges are all intraprocedural.
- detect nils in BasicBlock.Instrs.
- detect Function.Locals with Heap flag set.
- check fn.Blocks is nil if empty.
Also:
- Phi now has Comment field for debugging.
- Fixed bug in Select.Operands()
(took address of temporary copy of field)
- new Literal constructor zeroLiteral().
- algorithms steal private fields Alloc.index,
BasicBlock.gaps to avoid allocating maps.
- We print Function.Locals in DumpTo.
- added profiling support to ssadump.
R=iant, gri
CC=golang-dev
https://golang.org/cl/7229074
mpreinit() is called on the parent thread and with mcache (can allocate memory),
minit() is called on the child thread and can not allocate memory.
R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/7389043
Simplifies the contract for Driver.Stmt.Close in
the process of fixing issue 3865.
Fixes#3865
Update #4459 (maybe fixes it; uninvestigated)
R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/7363043
Add a new, simple interface for scanning (probably textual) data,
based on a new type called Scanner. It does its own internal buffering,
so should be plausibly efficient even without injecting a bufio.Reader.
The format of the input is defined by a "split function", by default
splitting into lines. Other implemented split functions include single
bytes, single runes, and space-separated words.
Here's the loop to scan stdin as a file of lines:
s := bufio.NewScanner(os.Stdin)
for s.Scan() {
fmt.Printf("%s\n", s.Bytes())
}
if s.Err() != nil {
log.Fatal(s.Err())
}
While we're dealing with spaces, define what space means to strings.Fields.
Fixes#4802.
R=adg, rogpeppe, bradfitz, rsc
CC=golang-dev
https://golang.org/cl/7322088
(Offsetof is a function of Alignof and Sizeof.)
- removed IntSize, PtrSize from Context (set Sizeof instead)
- GcImporter needs a Context now (it needs to have
access to Sizeof/Alignof)
- removed exported Size field from Basic (use Sizeof)
- added Offset to Field
- added Alignment, Size to Struct
R=adonovan
CC=golang-dev
https://golang.org/cl/7357046
The removed code leads to the situation when M executes the same locked G again
and again.
This is https://golang.org/cl/7310096 but with return instead of break
in the nested switch.
Fixes#4820.
R=golang-dev, alex.brainman, rsc
CC=golang-dev
https://golang.org/cl/7304102
On Windows, directory names in PATH can be fully or partially quoted
in double quotes ('"'), but the path names as used by most APIs must
be unquoted. In addition, quoted names can contain the semicolon
(';') character, which is otherwise used as ListSeparator.
This CL changes SplitList in path/filepath and LookPath in os/exec
to only treat unquoted semicolons as separators, and to unquote the
separated elements.
(In addition, fix harmless test bug I introduced for LookPath on Unix.)
Related discussion thread:
https://groups.google.com/d/msg/golang-nuts/PXCr10DsRb4/sawZBM7scYgJ
R=rsc, minux.ma, mccoyst, alex.brainman, iant
CC=golang-dev
https://golang.org/cl/7181047
The data file should be opened when a Conn is first
established, rather than waiting for the first Read or
Write.
Upon Close, we now make sure to try to close both, the
ctl as well as data files and set both to nil, even in
the face of errors, instead of returning early.
The Accept call was not setting the remote address
of the connection properly. Now, we read the correct
file.
Make functions that establish Conn use newTCPConn
or newUDPConn.
R=rsc, rminnich, ality, dave
CC=golang-dev
https://golang.org/cl/7228068
This CL changes nothing to existing API behavior, just sets up
Zone in IPNet and IPAddr structures if possible.
Also does small simplification.
Update #4234.
R=rsc, dave
CC=golang-dev
https://golang.org/cl/7300081
On Linux point-to-point interface an IFA_ADDRESS attribute
represents a peer address. For a correct interface address
we should take an IFA_LOCAL attribute instead.
Fixes#4839.
R=golang-dev, dave, rsc
CC=golang-dev
https://golang.org/cl/7352045
This avoids ambiguity and makes the diagnostics closer to
those issued by gc, but it is more verbose since it qualifies
intra-package references.
Without extra context---e.g. a 'from *Package' parameter to
Type.String()---we are forced to err on one side or the other.
Also, cosmetic changes to exp/ssa:
- Remove package-qualification workaround in Function.FullName.
- Always set go/types.Package.Path field to the import path,
since we know the correct path at this point.
- In Function.DumpTo, show variadic '...' and result type info,
and delete now-redundant "# Type: " line.
R=gri
CC=golang-dev
https://golang.org/cl/7325051
Also:
- faster code for example extraction
- simplify handling of command documentation:
all "main" packages are treated as commands
- various minor cleanups along the way
For commands written in Go, any doc.go file containing
documentation must now be part of package main (rather
then package documentation), otherwise the documentation
won't show up in godoc (it will still build, though).
For commands written in C, documentation may still be
in doc.go files defining package documentation, but the
recommended way is to explicitly ignore those files with
a +build ignore constraint to define package main.
Fixes#4806.
R=adg, rsc, dave, bradfitz
CC=golang-dev
https://golang.org/cl/7333046
If a test can be placed in the same package ("internal"), it is placed
there. This facilitates testing of package-private details. Because of
dependency cycles some packages cannot be tested by internal tests.
R=golang-dev, rsc, mikioh.mikioh
CC=golang-dev, r
https://golang.org/cl/7323044
The current implementation would store all cookies received from
any .com domain under "com" in the entries map if a nil public
suffix list is used in constructing the Jar. This is inefficient.
This CL uses the TLD+1 of the domain if the public suffix list
is nil which has two advantages:
- It uses the entries map efficiently.
- It prevents a host foo.com to set cookies for bar.com.
(It may set the cookie, but it won't be returned to bar.com.)
A domain like www.british-library.uk may still set a domain
cookie for .british-library.uk in this case.
The behavior for a non-nil public suffix list is unchanged, cookies
are stored under eTLD+1 in this case.
R=nigeltao
CC=golang-dev
https://golang.org/cl/7312105
* Handle p==nil in signalstack by setting SS_DISABLE flag.
* Make minit only allocate a signal g if there's not one already.
R=golang-dev, r
CC=golang-dev
https://golang.org/cl/7323072
Re-enable TestUpdateAndDelete, TestExpiration, TestChromiumDomain and
TestChromiumDeletion on Windows.
Sorting of cookies with same path length and same creation
time is done by an additional seqNum field.
This makes the order in which cookies are returned in Cookies
deterministic, even if the system clock is manipulated or on
systems with a low-resolution clock.
The tests now use a synthetic time: This makes cookie testing
reliable in case of bogus system clocks and speeds up the
expiration tests.
R=nigeltao, alex.brainman, dave
CC=golang-dev
https://golang.org/cl/7323063
Fix the sa_mask member of the sigaction struct - on FreeBSD this is
declared as a sigset_t, which is an array of four unsigned ints.
Replace the current int64 with Sigset from defs_freebsd_GOARCH, which
has the correct definition.
Unbreaks the FreeBSD builds.
R=golang-dev, dave, minux.ma
CC=golang-dev
https://golang.org/cl/7333047
broke windows build
««« original CL description
runtime: ensure forward progress of runtime.Gosched() for locked goroutines
The removed code leads to the situation when M executes the same locked G again and again.
Fixes#4820.
R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/7310096
»»»
TBR=dvyukov
CC=golang-dev
https://golang.org/cl/7343050
Arguably if this happens the program is buggy anyway,
but letting the panic continue looks better than interrupting it.
Otherwise things like this are possible, and confusing:
$ go run x.go
panic: $ echo $?
0
$
Fixes#3934.
R=golang-dev, r
CC=golang-dev
https://golang.org/cl/7322083
This is the same logic used in the standard tracebacks.
The caller pc is the pc after the call, so except in the
fake "call" caused by a panic, back up the pc enough
that the lookup will use the previous instruction.
Fixes#4150.
Fixes#4151.
R=golang-dev, iant
CC=golang-dev
https://golang.org/cl/7317047
Before, the mheap structure was in the bss,
but it's quite large (today, 256 MB, much of
which is never actually paged in), and it makes
Go binaries run afoul of exec-time bss size
limits on some BSD systems.
Fixes#4447.
R=golang-dev, dave, minux.ma, remyoudompheng, iant
CC=golang-dev
https://golang.org/cl/7307122
The removed code leads to the situation when M executes the same locked G again and again.
Fixes#4820.
R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/7310096
Right now it says 'invalid type S' for a struct type S.
Instead, say which type inside the struct is the problem.
Fixes#4825.
R=golang-dev, iant
CC=golang-dev
https://golang.org/cl/7301102
In addition to the compile failure fixed in signal*.c,
preserving the signal mask led to very strange crashes.
Testing shows that looking for SIG_IGN is all that
matters to get along with nohup, so reintroduce
sigset_zero instead of trying to preserve the signal mask.
TBR=iant
CC=golang-dev
https://golang.org/cl/7323067
There are two ways nohup(1) might be implemented:
it might mask away the signal, or it might set the handler
to SIG_IGN, both of which are inherited across fork+exec.
So two fixes:
* Make sure to preserve the inherited signal mask at
minit instead of clearing it.
* If the SIGHUP handler is SIG_IGN, leave it that way.
Fixes#4491.
R=golang-dev, mikioh.mikioh, iant
CC=golang-dev
https://golang.org/cl/7308102
Subject Alternative Names in X.509 certificates may include IP
addresses. This change adds support for marshaling, unmarshaling and
verifying this form of SAN.
It also causes IP addresses to only be checked against IP SANs,
rather than against hostnames as was previously the case. This
reflects RFC 6125.
Fixes#4658.
R=golang-dev, mikioh.mikioh, bradfitz
CC=golang-dev
https://golang.org/cl/7336046
It is too flaky. Tried to make it more reliable,
but that affects other tests (they run too long),
because we do unusual things here, like attempting
to connect to non-existing address and interrupt.
R=golang-dev, bradfitz, mikioh.mikioh
CC=golang-dev
https://golang.org/cl/7314097
Add support for arbitrary notes of the form // MARKER(userid): comment
in the same vein as BUG(userid): A marker must be two or more upper case [A-Z] letters.
R=gri, rsc, bradfitz, jscrockett01
CC=golang-dev
https://golang.org/cl/7322061
- use the new AllErrors flag where appropriate
- unless AllErrors is set, eliminate spurious
errors before they are added to the errors list
(it turns out that reporting spurious errors always
leads to too many uninformative errors after all)
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7323065
The second attempt at the Unmarshal optimization allowed
panics to get out of the json package. Add test for that bug
and remove the optimization.
Let's stop trying to optimize Unmarshal.
Fixes#4784.
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7300108
There wil be a panic if more than ten errors are encountered. ParseFile
will recover and return the ErrorList.
Fixes#3943.
R=golang-dev, gri
CC=golang-dev
https://golang.org/cl/7307085
Cleans up godoc and makes it consistent. (some had it, some
didn't)
This still keeps the information there, though, for people
looking at the source directly.
R=golang-dev, minux.ma, rsc
CC=golang-dev
https://golang.org/cl/7324056
Avoids the dot-dot-based algorithm on repeated calls
when the directory hasn't changed.
R=golang-dev, iant, bradfitz
CC=golang-dev
https://golang.org/cl/7340043
This CL provides the implementation of Cookies and
the complete test suite. Several tests have been ported
from the Chromium project as a cross check.
R=nigeltao, rsc, bradfitz
CC=golang-dev
https://golang.org/cl/7311073
No code changes.
This is mainly in preparation to scheduler changes,
oldstack/newstack are not related to scheduling.
R=golang-dev, minux.ma, rsc
CC=golang-dev
https://golang.org/cl/7311085
This is based on rsc's code posted to issue 2585.
Benchmark results are greatly improved:
benchmark old ns/op new ns/op delta
BenchmarkSortString1K 564397 445897 -21.00%
BenchmarkSortInt1K 270889 221249 -18.32%
BenchmarkSortInt64K 26850765 21351967 -20.48%
Eyeballing a sampling of the raw number of comparisons shows a drop
on the order of 20-30% almost everywhere. The test input data that
doesn't match that are some of sawtooth/rand/plateau distributions,
where there is no change in the number of comparisons; that is,
there are no situations where this makes *more* comparisons.
Fixes#2585.
R=iant, rsc
CC=golang-dev
https://golang.org/cl/7306098
Completly the same like the Execer-Interface, just for Queries.
This allows Drivers to execute Queries without preparing them first
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7085056
The test for issue 3590 causes an error to be printed to stderr when run (although the error is obscured during go test std). This is confusing for people who get breakage in the net package as the error is harmless and most likely unrelated to their build breakage.
Given the way the test works, by reaching into the guts of the netFD, I can't see a way to silence the error without adding a bunch of code to support the test, therefore I am suggesting the test be removed before Go 1.1 ships.
R=alex.brainman, mikioh.mikioh, rsc
CC=golang-dev
https://golang.org/cl/7307110
Changed accidentally in 28966b7b2f0c (CopyN using Copy).
Updating docs to be consistent with 29bf5ff5064e (ReadFull & ReadAtLeast)
R=rsc
CC=golang-dev
https://golang.org/cl/7314069
Unexported field and method names that appear in the
export data (as part of some exported type) are fully
qualified with a package id (path). In some cases, a
package with that id was never exported for any other
use (i.e. only the path is of interest).
We must not create a "real" package in those cases
because we don't have a package name. Entering an
unnamed package into the map of imported packages
makes that package accessible for other imports.
Such a subsequent import may find the unnamed
package in the map, and reuse it. That reused and
imported package is then entered into the importing
file scope, still w/o a name. References to that
package cannot resolved after that. Was bug.
R=adonovan
CC=golang-dev
https://golang.org/cl/7307112
Operands returns the SSA values used by an instruction.
Referrers returns the SSA instructions that use a value, for
some values. These will be used for SSA renaming, to follow.
R=iant, gri
CC=golang-dev
https://golang.org/cl/7312090
into separate package. This allows this code to be shared
with the search package without the need for these two to use
the same tables.
Adjusted various files accordingly.
R=rsc
CC=golang-dev
https://golang.org/cl/7213044
The lowering of ast.RangeStmt now has three distinct cases:
1) rangeIter for maps and strings; approximately:
it = range x
for {
k, v, ok = next it
if !ok { break }
...
}
The Range instruction and the interpreter's "iter"
datatype are now restricted to these types.
2) rangeChan for channels; approximately:
for {
k, ok = <-x
if !ok { break }
...
}
3) rangeIndexed for slices, arrays, and *array; approximately:
for k, l = 0, len(x); k < l; k++ {
v = x[k]
...
}
In all cases we now evaluate the side effects of the range expression
exactly once, per comments on http://code.google.com/p/go/issues/detail?id=4644.
However the exact spec wording is still being discussed in
https://golang.org/cl/7307083/. Further (small)
changes may be required once the dust settles.
R=iant
CC=golang-dev
https://golang.org/cl/7303074
This CL provides the rest of the SetCookies code as well as
some test infrastructure which will be used to test also
the Cookies method. This test infrastructure is optimized
for readability and tries to make it easy to review table
driven test cases.
Tests for all the different corner cases of SetCookies
will be provided in a separate CL.
R=nigeltao, rsc, bradfitz
CC=golang-dev
https://golang.org/cl/7306054
Support reading pax archives and applying extended attributes
like long names, subsecond mtime resolutions, etc. Default to
writing pax archives for long file and link names.
Support reading gnu archives using the ././@LongLink extended
header for file name and link target.
Fixes#3300
R=dsymonds, dave, rogpeppe, remyoudompheng, chressie, rsc
CC=golang-dev
https://golang.org/cl/6700047
On Windows, CryptoAPI is finding an alternative validation path. Since
this is a little non-deterministic, this change disables that test
when using system validation.
R=golang-dev
CC=golang-dev
https://golang.org/cl/7313068
By default, crypto/x509 assumes that users wish to validate
certificates for ServerAuth. However, due to historical reasons,
COMODO's intermediates don't specify ServerAuth as an allowed key
usage.
Rather NSS and CryptoAPI both allow these SGC OIDs to be equivalent to
ServerAuth.
R=rsc
CC=golang-dev
https://golang.org/cl/7312070
Add benchmark for request parsing. Fixture data is taken from https://github.com/felixge/node-http-perf
% go version
go version devel +28966b7b2f0c Thu Feb 07 20:26:12 2013 -0800 linux/amd64
% go test -run=nil -bench=ReadRequest -benchtime=10s
PASS
BenchmarkReadRequest 2000000 9900 ns/op 61.71 MB/s 3148 B/op 32 allocs/op
ok net/http 12.180s
R=golang-dev, bradfitz, minux.ma, haimuiba
CC=golang-dev
https://golang.org/cl/7313048
On POSIX, '=' in key is explicitly invalid, and '\x00' in key/value is implicitly invalid.
R=golang-dev, iant, bradfitz
CC=golang-dev
https://golang.org/cl/7311061
With the new scheduler races in the tests are reported during execution of other tests.
The change joins goroutines started during the tests.
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7310066
The problem happens when end=0, then end-1 is very big number.
Observed with the new scheduler.
R=golang-dev, iant
CC=golang-dev
https://golang.org/cl/7307073
It might be non-blocking, but it also might be blocking.
Cannot take the chance, as Accept might block indefinitely
and make it impossible to acquire ForkLock exclusively
(during fork+exec).
Fixes#4737.
R=golang-dev, dave, iant, mikioh.mikioh
CC=golang-dev
https://golang.org/cl/7309050
Before and after:
BenchmarkTrimSpace 20000000 81.3 ns/op
BenchmarkTrimSpace 50000000 58.0 ns/op
(most whitespace trimming is ASCII whitespace)
Same optimization appeared a handful of other places
in this file, but not here.
R=golang-dev, dave
CC=golang-dev
https://golang.org/cl/7305063
This is necessary for systems that use select as the pollster,
such as Solaris (supported by gccgo). It corresponds to the
bool returned by AddFD. In general it's not clearly defined
what happens when a descriptor used in a select is closed, and
different systems behave differently. Waking up the select
will cause the right thing to happen: the closed descriptor
will be dropped from the next iteration.
R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/7303056
deferred block. This makes hangs in the waiting code less likely
if a goroutine exits abnormally.
R=golang-dev, minux.ma
CC=golang-dev
https://golang.org/cl/7306052
This CL is the first of a handful of CLs which will provide
the implementation of cookiejar. It contains several helper
functions and the skeleton of Cookies and SetCookies.
Proper host name handling requires the ToASCII transformation
from package idna which currently lives in the go.net
subrepo. This CL thus contains just a TODO for this issue.
R=nigeltao, rsc, bradfitz
CC=golang-dev
https://golang.org/cl/7287046
Removes limit on maximum number of goroutines ever existed.
code.google.com/p/goexecutor tests now pass successfully.
Also slightly improves performance.
Before: $ time ./flate.test -test.short
real 0m9.314s
After: $ time ./flate.test -test.short
real 0m8.958s
Fixes#4286.
The runtime is built from llvm rev 174312.
R=rsc
CC=golang-dev
https://golang.org/cl/7218044
Closing the inotify file descriptor can take over a second
when running on Ubuntu Precise in an NFS directory, leading to
the test error in issue 3132. Closing the event channel first
lets a client that does not care about the error channel move
on.
Fixes#3132.
R=golang-dev, dave, rsc
CC=golang-dev
https://golang.org/cl/7300045
This updates a bad reference to a method name in the example priority queue test.
The error was introduced in the example refactoring in rev. 2ea8f07b2ffe.
R=golang-dev, minux.ma
CC=golang-dev
https://golang.org/cl/7279045
If a Handle's Write to a ResponseWriter fails (e.g. via a
net.Conn WriteDeadline via WriteTimeout on the Server), the
Server was blocking forever waiting for reads on that
net.Conn, even after a Write failed.
Instead, once we see a Write fail, close the connection,
since it's then dead to us anyway.
Fixes#4741
R=golang-dev, adg
CC=golang-dev
https://golang.org/cl/7301043
This provides the mechanism to connect SPDY support to the http
package, without pulling SPDY into the standard library.
R=rsc, agl, mikioh.mikioh
CC=golang-dev
https://golang.org/cl/7287045
This CL adds TCPInfo struct to linux/386,arm.
It's already added to linux/amd64.
Note that not sure the reason but cgo godefs w/ latest gcc
translates a flexible array member in structures correctly,
handles it as a non-incomplete, non-opaque type, on Go 1.
This CL reverts such changes by hand for the Go 1 contract.
R=minux.ma, bradfitz, rsc
CC=golang-dev
https://golang.org/cl/7197046
Deadlines should be extended at the beginning of
a request, not at the beginning of a connection.
Fixes#4676
R=golang-dev, fullung, patrick.allen.higgins, adg
CC=golang-dev
https://golang.org/cl/7220076
In cmd/go's 'go help testflag':
* Rewrite list of flags to drop test. prefix on every name.
* Sort list of flags.
* Add example of using -bench to match all benchmarks.
In testing:
* Remove mention of undefined 'CPU group' concept.
Fixes#4488.
Fixes#4508.
R=adg
CC=golang-dev
https://golang.org/cl/7288053
* Document Parse's zone interpretation.
* Add ParseInLocation (API change).
* Recognize "wrong" time zone names, like daylight savings time in winter.
* Disambiguate time zone names using offset (like winter EST vs summer EST in Sydney).
The final two are backwards-incompatible changes, but I believe
they are both buggy behavior in the Go 1.0 versions; the old results
were more wrong than the new ones.
Fixes#3604.
Fixes#3653.
Fixes#4001.
R=adg
CC=golang-dev
https://golang.org/cl/7288052
