There were at least three races in the implementation of the pool of
idle HTTP connections before this CL.
The first race is that HTTP/2 connections can be shared for many
requests, but each requesting goroutine would take the connection out
of the pool and then immediately return it before using it; this
created unnecessary, tiny little race windows during which another
goroutine might dial a second connection instead of reusing the first.
This CL changes the idle pool to just leave the HTTP/2 connection in
the pool permanently (until there is reason to close it), instead of
doing the take-it-out-put-it-back dance race.
The second race is that “is there an idle connection?” and
“register to wait for an idle connection” were implemented as two
separate steps, in different critical sections. So a client could end
up registered to wait for an idle connection and be waiting or perhaps
dialing, not having noticed the idle connection sitting in the pool
that arrived between the two steps.
The third race is that t.getIdleConnCh assumes that the inability to
send on the channel means the client doesn't need the result, when it
could mean that the client has not yet entered the select.
That is, the main dial does:
idleConnCh := t.getIdleConnCh(cm)
select {
case v := <-dialc:
...
case pc := <-idleConnCh
...
...
}
But then tryPutIdleConn does:
waitingDialer := t.idleConnCh[key] // what getIdleConnCh(cm) returned
select {
case waitingDialer <- pconn:
// We're done ...
return nil
default:
if waitingDialer != nil {
// They had populated this, but their dial won
// first, so we can clean up this map entry.
delete(t.idleConnCh, key)
}
}
If the client has returned from getIdleConnCh but not yet reached the
select, tryPutIdleConn will be unable to do the send, incorrectly
conclude that the client does not care anymore, and put the connection
in the idle pool instead, again leaving the client dialing unnecessarily
while a connection sits in the idle pool.
(It's also odd that the success case does not clean up the map entry,
and also that the map has room for only a single waiting goroutine for
a given host.)
None of these races mattered too much before Go 1.11: at most they
meant that connections were not reused quite as promptly as possible,
or a few more than necessary would be created. But Go 1.11 added
Transport.MaxConnsPerHost, which limited the number of connections
created for a given host. The default is 0 (unlimited), but if a user
did explicitly impose a low limit (2 is common), all these misplaced
conns could easily add up to the entire limit, causing a deadlock.
This was causing intermittent timeouts in TestTransportMaxConnsPerHost.
The addition of the MaxConnsPerHost support added its own races.
For example, here t.incHostConnCount could increment the count
and return a channel ready for receiving, and then the client would
not receive from it nor ever issue the decrement, because the select
need not evaluate these two cases in order:
select {
case <-t.incHostConnCount(cmKey):
// count below conn per host limit; proceed
case pc := <-t.getIdleConnCh(cm):
if trace != nil && trace.GotConn != nil {
trace.GotConn(httptrace.GotConnInfo{Conn: pc.conn, Reused: pc.isReused()})
}
return pc, nil
...
}
Obviously, unmatched increments are another way to get to a deadlock.
TestTransportMaxConnsPerHost deadlocked approximately 100% of
the time with a small random sleep added between incHostConnCount
and the select:
ch := t.incHostConnCount(cmKey):
time.Sleep(time.Duration(rand.Intn(10))*time.Millisecond)
select {
case <-ch
// count below conn per host limit; proceed
case pc := <-t.getIdleConnCh(cm):
...
}
The limit also did not properly apply to HTTP/2, because of the
decrement being attached to the underlying net.Conn.Close
and net/http not having access to the underlying HTTP/2 conn.
The alternate decrements for HTTP/2 may also have introduced
spurious decrements (discussion in #29889). Perhaps those
spurious decrements or other races caused the other intermittent
non-deadlock failures in TestTransportMaxConnsPerHost,
in which the HTTP/2 phase created too many connections (#31982).
This CL replaces the buggy, racy code with new code that is hopefully
neither buggy nor racy.
Fixes#29889.
Fixes#31982.
Fixes#32336.
Change-Id: I0dfac3a6fe8a6cdf5f0853722781fe2ec071ac97
Reviewed-on: https://go-review.googlesource.com/c/go/+/184262
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Test case that causes incorrect compiler error from gccgo.
Updates #32922
Change-Id: I59432a8e8770cf03eda293f6d110c081c18fa88b
Reviewed-on: https://go-review.googlesource.com/c/go/+/184918
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
It wasn't obeyed in the case where the MarshalJSON method uses a pointer
receiver, and the encoder grabs the address of a value to find that
method. addrMarshalerEncoder is the function that does this work, but it
ignored opts.escapeHTML.
Here's the before and after of the added test case, which was failing
before the fix. Now the two cases are correct and consistent.
{"NonPtr":"<str>","Ptr":"\u003cstr\u003e"}
{"NonPtr":"<str>","Ptr":"<str>"}
Fixes#32896.
Change-Id: Idc53077ece074973558bd3bb5ad036380db0d02c
Reviewed-on: https://go-review.googlesource.com/c/go/+/184757
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
Reviewed-by: Caleb Spare <cespare@gmail.com>
cgo produces dynamic imports for Go binaries by scanning the dynamic
imports table of a binary produced by the system C compiler and
linker. Currently, since it uses elf.File.ImportedSymbols, it only
reads global symbols. Unfortunately, recent versions of lld emit weak
symbol imports for several pthread symbols, which means the cgo tool
doesn't emit dynamic imports for them, which ultimately causes linking
of cgo binaries to fail.
Fix this by using elf.File.DynamicSymbols instead and filtering down
to both global and weak symbols.
Fixes#31912.
Change-Id: If346a7eca6733e3bfa2cccf74a9cda02a3e81d38
Reviewed-on: https://go-review.googlesource.com/c/go/+/184100
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Currently, File.ImportedSymbols is the only API that exposes the GNU
symbol version information for dynamic symbols. Unfortunately, it also
filters to specific types of symbols, and only returns symbol names.
The cgo tool is going to need symbol version information for more
symbols. In order to support this and make the API more orthogonal,
this CL adds version information to the Symbol type and updates
File.DynamicSymbols to fill this in. This has the downside of
increasing the size of Symbol, but seems to be the most natural API
for exposing this. I also explored 1) adding a method to get the
version information for the i'th dynamic symbol, but we don't use
symbol indexes anywhere else in the API, and it's not clear if this
index would be 0-based or 1-based, and 2) adding a
DynamicSymbolVersions method that returns a slice of version
information that parallels the DynamicSymbols slice, but that's less
efficient to implement and harder to use.
For #31912.
Change-Id: I69052ac3894f7af2aa9561f7085275130e0cf717
Reviewed-on: https://go-review.googlesource.com/c/go/+/184099
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
- fix link for `Time.Format`
- fix closing tag for `go get`
- add links for `runtime.Caller`, `runtime.Callers`
- remove link for `TypedArrayOf` since it has been removed (CL 177537)
Change-Id: I1dc38226e6d91c68fbd2f02c1acfad5327f4ebe8
Reviewed-on: https://go-review.googlesource.com/c/go/+/185038
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
Running the example code when not having permissions
to bind to port 80 will cause the program to hang after
printing the error message.
Change-Id: I2433ba2629b362fc8f1731e40cab5eea72ec354f
GitHub-Last-Rev: 0bb3dc08b6
GitHub-Pull-Request: golang/go#32947
Reviewed-on: https://go-review.googlesource.com/c/go/+/185157
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This should have been part of CL 184440.
Updates #32846
Change-Id: I78a1326f4a67b3b526859bd15cb9653b4a8551a7
Reviewed-on: https://go-review.googlesource.com/c/go/+/184920
Run-TryBot: Jay Conrod <jayconrod@google.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
'go get path@latest' may now downgrade a module required at a
pre-release or pseudo-version newer than the latest released
version. This restores the 1.12 behavior and the ability to easily
roll back from a temporary development version.
'go get path@upgrade' is like @latest but will not downgrade.
If no version suffix is specified ('go get path'), @upgrade is
implied.
Fixes#32846
Change-Id: Ibec0628292ab1c484716a5add0950d7a7ee45f47
Reviewed-on: https://go-review.googlesource.com/c/go/+/184440
Run-TryBot: Jay Conrod <jayconrod@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
The documentation just said "alphanumeric", but underscores and dots
are also accepted.
Fixes#32886
Change-Id: I1ba872a220d5c5bf64f1d851ddba9eb3b1afb89a
Reviewed-on: https://go-review.googlesource.com/c/go/+/184917
Reviewed-by: Bryan C. Mills <bcmills@google.com>
This CL adds a test for gccgo bug #32901: not all the type
descriptors are registered and thus deduplicated with types
created by reflection. It needs a few levels of indirect imports
to trigger this bug.
Updates #32901.
Change-Id: Idbd89bedd63fea746769f2687f3f31c9767e5ec0
Reviewed-on: https://go-review.googlesource.com/c/go/+/184718
Reviewed-by: Ian Lance Taylor <iant@golang.org>
I accidentally fetched an invalid version of rsc.io/quote from
proxy.golang.org, which the proxy then cached and now includes in
https://proxy.golang.org/rsc.io/quote/@v/list.
That causes 'go get rsc.io/quote` to resolve to a different version
depending on whether the proxy is used.
Adjust the test to fetch an explicit version instead, since the choice
of 'latest' is mostly irrelevant to the checksum database logic that
the test is intended to verify.
Updates #32805Fixes#32900
Change-Id: I075b1f62e8c71545d0fb2dd4bd77ba525fc2a36d
Reviewed-on: https://go-review.googlesource.com/c/go/+/184719
Run-TryBot: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Test case that caused a compiler crash in gofrontend, related to
exporting inlinable function bodies.
Updates #32778
Change-Id: Iacf1753825d5359da43e5e281189876d4c3dd3c9
Reviewed-on: https://go-review.googlesource.com/c/go/+/183851
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Taking over CL 162240, the original CL hasn't been making progress.
I just took the parts that fix the immediate issue. I left the
signatslice changes out, I don't think they are necessary.
Fixes#30202
Change-Id: I5b347605f0841dd925d5a73150b8bf269fa82464
Reviewed-on: https://go-review.googlesource.com/c/go/+/183852
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
It's not correct to use atomic.CompareAndSwap to implement Once.Do,
and we don't, but why we don't is a question that has come up
twice on golang-dev in the past few months.
Add a comment to help others with the same question.
Change-Id: Ia89ec9715cc5442c6e7f13e57a49c6cfe664d32c
Reviewed-on: https://go-review.googlesource.com/c/go/+/184261
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Ingo Oeser <nightlyone@googlemail.com>
The implementation of semaphores, and therefore notes, used on Darwin
is not async-signal-safe. The runtime has one case where a note needs
to be woken up from a signal handler: the call to notewakeup in sigsend.
That notewakeup call is only called on a single note, and it doesn't
need the full functionality of notes: nothing ever does a timed wait on it.
So change that one note to use a different implementation on Darwin,
based on a pipe. This lets the wakeup code use the write call, which is
async-signal-safe.
Fixes#31264
Change-Id: If705072d7a961dd908ea9d639c8d12b222c64806
Reviewed-on: https://go-review.googlesource.com/c/go/+/184169
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
The localPipe implementation assumes that every successful net.Dial
results in exactly one successful listener.Accept. I don't believe this
is guaranteed by essentially any operating system. For this test, we're
seeing flakes on dragonfly (#29583).
But see also #19519, flakes due to the same assumption on FreeBSD
and macOS in package net's own tests.
This CL rewrites localPipe to try a few times to get a matching pair
of connections on the dial and accept side.
Fixes#29583.
Change-Id: Idb045b18c404eae457f091df20456c5ae879a291
Reviewed-on: https://go-review.googlesource.com/c/go/+/184157
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
TestVariousDeadlines starts a client and server.
The client dials the server, sets a timeout on the connection,
reads from it, gets a timeout error, closes the connection.
The server writes an infinite stream of a's to each connection
it accepts.
The test was trying to run these in lockstep:
run a client dial+read+timeout+close,
wait for server to accept+write+error out on write to closed connection,
repeat.
On FreeBSD 11.2 and less frequently on macOS we see
the test timeout waiting for the server to do its half of
the lockstep dance.
I believe the problem is that the client can do its step
of the dance with such a short timeout that the read,
timeout, and close happens before the server ever returns
from the accept(2) system call. For the purposes of testing
the client-side read timeout, this is fine. But I suspect
that under some circumstances, the "TCP-accepted"
connection does not translate into a "socket-layer-accepted"
connection that triggers a return from accept(2).
That is, the Go server never sees the connection at all.
And the test sits there waiting for it to acknowledge
being done with a connection it never started with.
Fix the problem by not trying to lockstep with the server.
This definitely fixes the flake, since the specific line that
was calling t.Fatal is now deleted.
This exposes a different flake, seen on a trybot run for an
early version of this CL, in which the client's io.Copy does
not stop within the time allotted. The problem now is that
there is no guarantee that a read beyond the deadline with
available data returns an error instead of the available data,
yet the test assumes this guarantee, and in fact the opposite
is usually true - we don't bother checking the deadline unless
the read needs to block. That is, deadlines don't cut off a
flood of available data, yet this test thinks they do.
This CL therefore also changes the server not to send an
infinite flood of data - don't send any data at all - so that
the read deadline is guaranteed to be exercised.
Fixes#19519.
Change-Id: I58057c3ed94ac2aebab140ea597f317abae6e65e
Reviewed-on: https://go-review.googlesource.com/c/go/+/184137
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
CL 42652 changed the profile handler for mips/mipsle to
avoid recording a profile when in atomic functions, for fear
of interrupting the 32-bit simulation of a 64-bit atomic with
a lock. The profile logger itself uses 64-bit atomics and might
deadlock (#20146).
The change was to accumulate a count of dropped profile events
and then send the count when the next ordinary event was sent:
if prof.hz != 0 {
+ if (GOARCH == "mips" || GOARCH == "mipsle") && lostAtomic64Count > 0 {
+ cpuprof.addLostAtomic64(lostAtomic64Count)
+ lostAtomic64Count = 0
+ }
cpuprof.add(gp, stk[:n])
}
CL 117057 extended this behavior to include GOARCH == "arm".
Unfortunately, the inserted cpuprof.addLostAtomic64 differs from
the original cpuprof.add in that it neglects to acquire the lock
protecting the profile buffer.
This has caused a steady stream of flakes on the arm builders
for the past 12 months, ever since CL 117057 landed.
This CL moves the lostAtomic count into the profile buffer and
then lets the existing addExtra calls take care of it, instead of
duplicating the locking logic.
Fixes#24991.
Change-Id: Ia386c40034fcf46b31f080ce18f2420df4bb8004
Reviewed-on: https://go-review.googlesource.com/c/go/+/184164
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
The current cmd/doc implementation uses go/build.Import in a few
places to check whether a package is findable and importable.
go/build has limited support for finding packages in modules,
but to do so, build.Import requires knowing the source directory
to use when performing the lookup (so it can find the go.mod file).
Otherwise, it only looks inside the GOPATH workspace.
Start passing the current working directory to build.Import calls,
so that it can correctly look for packages in modules when in cmd/doc
is executed in module mode.
Before this change, cmd/doc in module mode could mistakenly find and
use a package in the GOPATH workspace, instead of the current module.
Since the result of os.Getwd is needed in even more places, assign it
to a local variable in parseArgs now.
Fixes#28992
Updates #26504
Change-Id: I7571618e18420d2d3b3890cc69ade2d97b1962bf
Reviewed-on: https://go-review.googlesource.com/c/go/+/183991
Run-TryBot: Dmitri Shuralyov <dmitshur@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
When reading tip.golang.org/doc/go1.13.html, the spec links in
the "Changes to the language" section should point to the updated
spec, not the old one.
Change-Id: I6b13ca0b4c722ed52b84a12a680bece876a4e478
Reviewed-on: https://go-review.googlesource.com/c/go/+/184118
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
The scanner was changed to accept the new Go number literal syntax
of which separators are a part. Making them opt-in is inconsistent
with the rest of the changes. For comparison, the strconv package
also accepts the new number literals including separators with the
various conversion routines, if no explicit number base is given.
Updates #28493.
Change-Id: Ifaae2225a9565364610813658bfe692901dd3ccd
Reviewed-on: https://go-review.googlesource.com/c/go/+/184080
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
go/build.Context.Import loads package information using 'go list' when
in module mode. It does this when GO111MODULE is not "off", there is a
go.mod file in any parent directory, and neither the path nor the
source directory are in GOROOT. Import no longer checks whether the
source directory is in GOPATH if GO111MODULE=auto or unset.
Also fixed subdirectory checks that did not handle relative source
directory paths. mod_gobuild_import should have failed when we changed
the meaning of GO111MODULE=auto but didn't because of this.
Fixes#32799
Change-Id: Ic5210b7e00cb58f91ea9455b67b49d5aed4eec63
Reviewed-on: https://go-review.googlesource.com/c/go/+/184098
Run-TryBot: Jay Conrod <jayconrod@google.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Put the sumdb cache in the usual pkg/mod/cache/download dir, rather
than the new pkg/mod/download/cache dir which I presume was a typo.
Change-Id: Id162f24db30f509353178ca0c8cc7a4dabc927e1
Reviewed-on: https://go-review.googlesource.com/c/go/+/183318
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Jay Conrod <jayconrod@google.com>
Looking at the live release notes on tip.golang.org, the Modules
section is much more verbose than the other sections.
To some extent that's to be expected, but too much detail in the
release notes might discourage folks from consulting the actual
documentation. Ensure that topics have clear links and omit
unnecessary details.
Change-Id: I1ccbc1697fccaf7ca7094c606bd11696c46d87f0
Reviewed-on: https://go-review.googlesource.com/c/go/+/183987
Run-TryBot: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Jay Conrod <jayconrod@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
In TestPoolDequeue it's surprisingly common for the queue to stay
nearly empty the whole time and for a racing PopTail to happen in the
window between the producer doing a PushHead and doing a PopHead. In
short mode, there are only 100 PopTail attempts. On linux/amd64, it's
not uncommon for this to fail 50% of the time. On linux/arm64, it's
not uncommon for this to fail 100% of the time, causing the test to
fail.
This CL fixes this by only checking for a successful PopTail in long
mode. Long mode makes 200,000 PopTail attempts, and has never been
observed to fail.
Fixes#31422.
Change-Id: If464d55eb94fcb0b8d78fbc441d35be9f056a290
Reviewed-on: https://go-review.googlesource.com/c/go/+/183981
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
TestPoolDequeue creates P-1 consumer goroutines and 1 producer
goroutine. Currently, if a consumer goroutine pops the last value from
the dequeue, it sets a flag that stops all consumers, but the producer
also periodically pops from the dequeue and doesn't set this flag.
Hence, if the producer were to pop the last element, the consumers
will continue to run and the test won't terminate. This CL fixes this
by also setting the termination flag in the producer.
I believe it's impossible for this to happen right now because the
producer only pops after pushing an element for which j%10==0 and the
last element is either 999 or 1999999, which means it should never try
to pop after pushing the last element. However, we shouldn't depend on
this reasoning.
Change-Id: Icd2bc8d7cb9a79ebfcec99e367c8a2ba76e027d8
Reviewed-on: https://go-review.googlesource.com/c/go/+/183980
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
TestPoolDequeue in long mode does a little more than 1<<21 pushes.
This was originally because the head and tail indexes were 21 bits and
the intent was to test wrap-around behavior. However, in the final
version they were both 32 bits, so the test no longer tested
wrap-around.
It would take too long to reach 32-bit wrap around in a test, so
instead we initialize the poolDequeue with indexes that are already
nearly at their limit. This keeps the knowledge of the maximum index
in one place, and lets us test wrap-around even in short mode.
Change-Id: Ib9b8d85b6d9b5be235849c2b32e81c809e806579
Reviewed-on: https://go-review.googlesource.com/c/go/+/183979
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
The @latest proxy endpoint is optional. If a proxy returns a 404 for
it, and returns an @v/list with no matching versions, then we should
allow module lookup to try other module paths. However, if the proxy
returns some other error (say, a 403 or 505), then the result of the
lookup is ambiguous, and we should report the actual error rather than
"no matching versions for query".
(This fix was prompted by discussion with Dmitri on CL 183619.)
Updates #32715
Updates #26334
Change-Id: I6d510a5ac24d48d9bc5037c3c747ac50695c663f
Reviewed-on: https://go-review.googlesource.com/c/go/+/183845
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
The test currently usses llvm.org, which seems to be very flaky today.
Change-Id: I3d01476d53f94d9170dbb087e3f3cf99581cdb4d
Reviewed-on: https://go-review.googlesource.com/c/go/+/183847
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
As suggested by thepudds in CL 183630.
Also adjust the paragraph to harmonize the transitions between the
newly-adjacent paragraphs.
Change-Id: Ie85abea946db81804c1995d27be4951d5db6b812
Reviewed-on: https://go-review.googlesource.com/c/go/+/183918
Reviewed-by: Jay Conrod <jayconrod@google.com>
Currently we use R16 and R17 for ARM64's Duff's devices.
According to ARM64 ABI, R16 and R17 can be used by the (external)
linker as scratch registers in trampolines. So don't use these
registers to pass information across functions.
It seems unlikely that calling Duff's devices would need a
trampoline in normal cases. But it could happen if the call
target is out of the 128 MB direct jump limit.
The choice of R20 and R21 is kind of arbitrary. The register
allocator allocates from low-numbered registers. High numbered
registers are chosen so it is unlikely to hold a live value and
forces a spill.
Fixes#32773.
Change-Id: Id22d555b5afeadd4efcf62797d1580d641c39218
Reviewed-on: https://go-review.googlesource.com/c/go/+/183842
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
When 'go install' is run without arguments in a directory containing a
main package, it deletes an executable file with the same name as the
package (presumably created by 'go build' previously).
'go get' in module mode executes the same code after updating and
downloading modules. However, the special case was misfiring because
we passed an empty list of patterns to InstallPackages.
Fixes#32766
Change-Id: I19aca64ee1fb5a216777dd7d559e8e6a45b3e90c
Reviewed-on: https://go-review.googlesource.com/c/go/+/183846
Run-TryBot: Jay Conrod <jayconrod@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Don't attempt to dereference basep if it's nil, just pass it to
getdirentries_freebsd12 as is.
Ported from x/sys/unix CL 183223
Change-Id: Id1c4e0eb6ff36dd39524da8194fed9a5957bce61
Reviewed-on: https://go-review.googlesource.com/c/go/+/183797
Reviewed-by: Ian Lance Taylor <iant@golang.org>
It would be nice to have a test, but it requires running
this under the race detector which is a bit complicated
to set up; yet the fix is trivial. Verified manually that
it doesn't trip the race detector.
Fixes#32154.
Change-Id: I20bd746a07945c802f0476a1d8b1dfd83c87dae8
Reviewed-on: https://go-review.googlesource.com/c/go/+/183849
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
It ends up making two similar types, [N]uint8 of both
alg and noalg varieties. Comparsions between the two then
don't come out equal when they should.
In particular, the type *[N]uint8 has an Elem pointer which
must point to one of the above two types; it can't point to both.
Thus allocating a *[N]uint8 and dereferencing it might be a
different type than a [N]uint8.
The fix is easy. Making a small test for this is really hard. It
requires that both a argless defer and the test be imported by a
common parent package. This is why a main binary doesn't see this
issue, but a test does (as Agniva noticed), because there's a wrapper
package that imports both the test and the defer.
Types like [N]uint8 don't really need to be marked noalg anyway,
as the generated code (if any) will be shared among all
vanilla memory types of the same size.
Fixes#32595
Change-Id: If7b77fa6ed56cd4495601c3f90170682d853b82f
Reviewed-on: https://go-review.googlesource.com/c/go/+/182357
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
