Roughly 33% faster for simple cases, probably more for complex ones.
Before:
mallocs per Sprintf(""): 4
mallocs per Sprintf("xxx"): 6
mallocs per Sprintf("%x"): 10
mallocs per Sprintf("%x %x"): 12
Now:
mallocs per Sprintf(""): 2
mallocs per Sprintf("xxx"): 3
mallocs per Sprintf("%x"): 5
mallocs per Sprintf("%x %x"): 7
Speed improves because of avoiding mallocs and also by sharing a bytes.Buffer
between print.go and format.go rather than copying the data back after each
printed item.
Before:
fmt_test.BenchmarkSprintfEmpty 1000000 1346 ns/op
fmt_test.BenchmarkSprintfString 500000 3461 ns/op
fmt_test.BenchmarkSprintfInt 500000 3671 ns/op
Now:
fmt_test.BenchmarkSprintfEmpty 2000000 995 ns/op
fmt_test.BenchmarkSprintfString 1000000 2745 ns/op
fmt_test.BenchmarkSprintfInt 1000000 2391 ns/op
fmt_test.BenchmarkSprintfIntInt 500000 3751 ns/op
I believe there is more to get but this is a good milestone.
R=rsc
CC=golang-dev, hong
https://golang.org/cl/166076
For 386 we use the [f]statfs64 system call, which takes three
parameters: the filename, the size of the statfs64 structure,
and a pointer to the structure itself.
R=rsc
https://golang.org/cl/166073
On a microbenchmark that ping-pongs on lots of channels, this makes
the multithreaded case about 20% faster and the uniprocessor case
about 1% slower. (Due to cache effects, I expect.)
R=rsc, agl
CC=golang-dev
https://golang.org/cl/166043
Use them in Copy and Copyn.
Speed up ReadFile by using ReadFrom and avoiding Copy altogether (a minor win).
R=rsc, gri
CC=golang-dev
https://golang.org/cl/166041
tabs for indentation even if -spaces is set.
Changes to gofmt:
- added -tabindent flag
- don't recompute parser and printer mode repeatedly
Changes to go/printer:
- provide new printing mode TabIndent
Changes to tabwriter:
- implement new mode TabIndent to use tabs independent
of the actual padding character for leading empty columns
- distinguish between minimal cell width and tab width
(tabwidth is only used if the output contains tabs,
minwidth and padding are always considered)
- fixed and added more comments
- some additional factoring
By default, -tabindent is disabled and the default gofmt
behavior is unchanged. By setting -spaces and -tabindent,
gofmt will use tabs for indentation but do any other
alignment with spaces. This permits a user to change the
visible indentation by simply changing the editor's tab
width and the code will remain properly aligned without
the need to rerun gofmt.
R=rsc
https://golang.org/cl/163068
1) need to send slice and array types (was only sending element types)
2) compatibleType needs to use decoder's type map
R=rsc
CC=golang-dev
https://golang.org/cl/164062
The python script needs a checkout of xcb/proto to generate
an xproto.go file, which together with xgb.go provide functions
to access all of the core X11 protocol requests. I have included the
generated file.
Extensions and authentication methods are not implemented.
R=r, rsc, nigeltao_golang
https://golang.org/cl/162053
this is the exact same thing issue #115 is about. fix makefiles to use relative
path to work in the case we have whitespaces as part of GOROOT.
R=rsc
https://golang.org/cl/162055
This provides an experimental X11 backend for the exp/draw interface.
It does not aim to provide a complete implementation of the X11 client protocol.
This works for me (Ubuntu Hardy 8.04, GOARCH=386). Your mileage my vary.
R=r, rsc, r1
CC=golang-dev
https://golang.org/cl/156109
before this change, if pkg/Make.deps is missing or broken, clean.bash fails and the build dies
but not until much later.
add freebsd to error message about valid values of $GOOS
TODO: would be nice if this process exited when an error occurred. subshells make it hard
R=rsc
CC=golang-dev
https://golang.org/cl/160065
the bash scripts and makefiles for building go didn't take into account
the fact $GOROOT / $GOBIN could both be directories containing whitespaces,
and was not possible to build it in such a situation.
this commit adjusts the various makefiles/scripts to make it aware of that
possibility, and now it builds successfully when using a path with whitespaces
as well.
Fixes#115.
R=rsc, dsymonds1
https://golang.org/cl/157067
a little slow, but usable (speed unchanged when not using -r)
tweak go/printer to handle nodes without line numbers
more gracefully in a couple cases.
R=gri
https://golang.org/cl/156103
Meant as illustration of the Go pattern that is using
goroutines and channels to handle exceptional situations.
Note: There is no need for "Finally" since the
"try block" (the function f supplied to Try)
cannot do a Smalltalk-style non-local return
and terminate the function surrounding Try.
Replaces CL 157083.
R=r, rsc
https://golang.org/cl/157087
* add runtime sliceslice1 for x[lo:]
* remove runtime arraytoslice, rewriting &arr into arr[0:len(arr)].
* port cgen_inline into 8g, 5g.
* use native memmove in maps
R=ken2
https://golang.org/cl/157106
Allows the developer to pass a map either by itself for
evaluation, or inside a struct. Access to data inside
maps is identical to the current system for structs, ie.
-Psuedocode-
mp map[string]string = {
"header" : "A fantastic header!",
"footer" : "A not-so-fantastic footer!",
}
template.Execute(mp)
...can be accessed using {header} and {footer} in
the template. Similarly, for maps inside structs:
type s struct {
mp map[string]string,
}
s1 = new s
s1.mp["header"] = "A fantastic header!";
template.Execute(s1)
...is accessed using {mp.header}. Multi-maps, ie.
map[string](map[string]string) and maps of structs
containing more maps are unsupported, but then, I'm
not even sure if that's supported by the language.
Map elements can be of any type that can be written
by the formatters. Keys should really only be strings.
Fixes#259.
R=r, rsc
https://golang.org/cl/157088
There's no functional change here. io gives the Read and Write methods byte slice arguments, but tar called them uint8. It's the same thing, but I think this is clearer and it matches what other packages do.
R=rsc
CC=golang-dev
https://golang.org/cl/157099
1) if char class contains a single character, make it a single character.
(this is used to quote, e.g. [.] rather than \.
2) if regexp begins with ordinary text substring, use plain string match to start engine
R=rsc
CC=golang-dev
https://golang.org/cl/157095
No benchmarks are run unless the --benchmarks=<regexp> flag
is specified on the gotest command line. This change includes
sample benchmarks for regexp.
% gotest --benchmarks=.*
(standard test output redacted)
testing.BenchmarkSimpleMatch 200000 7799 ns/op
testing.BenchmarkUngroupedMatch 20000 76898 ns/op
testing.BenchmarkGroupedMatch 50000 38148 ns/op
R=r, rsc
https://golang.org/cl/154173
also pick off the special case in strings.Index. don't want strings.IndexByte
because the call site will very rarely need to allocate and we can handle the
test in the code itself. bytes.IndexByte can avoid a common allocation.
R=rsc
CC=golang-dev
https://golang.org/cl/156091
Previously a netFd could be queued for reading/writing in the channel,
but close(2)'ed before pollServer got to it. In this case, the kernel
would consider the descriptor closed and the attempt to add it to the
epoll set would fail and panic.
This patch makes Close a roundtrip to the pollServer, although the
actual close(2) still occurs elsewhere to avoid blocking the
pollServer.
Fixes#143.
R=rsc
CC=golang-dev
https://golang.org/cl/152130
7x speedup on big and crypto/rsa unit tests.
also dropped useAsm in favor of making the
asm stubs jump to the Go versions.
R=agl1
CC=golang-dev, gri
https://golang.org/cl/157062
In thread.c, we need to cast to whatever the native
size of intptr is on the system, but we only have
uintptr available. They're the same size, but can't
do signed casts without this one :).
R=rsc
CC=golang-dev
https://golang.org/cl/156073
* move memmove to arch-specific subdirectories
* add memmove for arm
* add copyright notices marking them as copied from Inferno
R=ken2
https://golang.org/cl/156061
* add Marshal
* add BitString.RightAlign
* change to using a *time.Time (from time.Time) since that's what
the time package uses.
* return the unparsed data from Unmarshal.
R=rsc
CC=golang-dev
https://golang.org/cl/156047
the signal handling stack is a different size than
the normal stack, so it cannot be allocated using
the backup stack allocator.
Fixes#250.
R=agl1
CC=golang-dev
https://golang.org/cl/157044
Error information is carried from RPC server to client in the string
'Error' field of rpc.Response. An empty string is sent in the success
case. This empty string was being returned to the caller (of Client.Call
or Client.Go), resulting in a non-nil error response.
This change detects an empty-string Response.Error at the client, and
translates it into a nil value in Call.Error.
Tests updated to check error return in success cases.
R=r, rsc
https://golang.org/cl/154159
cgo/libmach remain unimplemented. However, compilers, runtime,
and packages are 100%. I still need to go through and implement
missing syscalls (at least make sure they're all listed), but
for all shipped functionality, this is done. Ship! ;)
R=rsc, VenkateshSrinivas
https://golang.org/cl/152142
We are dealing with the multiplicative group ℤ/pqℤ. Multiples of
either p or q are not members of the group since they cannot have an
inverse. (Such numbers are 0 in the subgroup ℤ/pℤ.)
With p and q of typical size (> 512 bits), the probability of a random
blind [1..pq-1] being a multiple of p or q is negligible. However, in
the unit tests, much smaller sizes are used and the event could occur.
This change checks the result of the ext GCD and deals with this case.
It also increases the size of p and q in the unit test as a large
number of the keys selected were p, q = 227,169.
R=rsc
CC=golang-dev
https://golang.org/cl/154141
The ByteOrder.Put* methods are already available, this change uses
them to implement the Write function.
R=golang-dev, agl1, rsc, r
https://golang.org/cl/152141
performance hit of about 20% but more intuitive results for submatches.
we need a good regexp package at some point.
Fixes#110.
R=rsc
CC=golang-dev
https://golang.org/cl/152131
Commented both flush methods so people know what they are looking at.
This is a necessary fix for streaming and long polling HTTP services.
Fixes#93.
R=r, rsc, david.titarenco
https://golang.org/cl/154099
SELinux will cause mmap to fail when we request w+x memory unless the
user has configured their policies. We have a warning in make.bash,
but it's quite likely that the policy will be reset at some point and
then all their binaries start failing.
This patch prints a warning on Linux when mmap fails with EACCES.
R=rsc
CC=golang-dev
https://golang.org/cl/152086
Turn methods that don't store the result in their receiver into
functions in order to preserve the convention.
Re-jig Exp and Div by moving their guts into nat.go.
Add ProbablyPrime to perform Miller-Rabin primality tests.
crypto/rsa: reenable key generation since we now have ProbablyPrime.
R=gri
CC=go-dev
http://codereview.prom.corp.google.com/1024038
if suffixes don't work, check for name directly.
also fixes short names like bit.ly when ndots>1.
tested by tossing domain and search lines from /etc/resolv.conf
Fixes#2.
R=agl, agl1
CC=golang-dev
https://golang.org/cl/152048
a node fits on one line:
- for purposes of measuring the node size in text,
don't generate html or use a styler that could
generate html as it will lead to overly large
sizes
A consequence of this bug is that source code displayed
with godoc may show functions that fit on one line in
the source on multiple lines.
This change causes no difference to the gofmt formatting
of any files in src or misc.
R=rsc
http://go/go-review/1026034
HTML vs Html, URL vs Url, HTTP vs Http,
current source is 6:1 in favor of the former,
so change instances of the latter.
R=r
CC=go-dev
http://go/go-review/1024026
reflect is a little more detailed than some because it affords an opportunity
to explain how to approach the library.
R=gri, rsc
CC=go-dev
http://go/go-review/1026026
- blank padding around composite literal contents for a less dense look
(most if not all composite literals were written in that style before gofmt
ran through them)
- corresponding (internal) flag: compositeLitBlank
- don't print a semi after a one-statement statement list (as preferred by r)
- corresponding (internal flag): fewerSemis
- the number of changes in nodes.go is huge because of the removed semis;
hg mail requires that I gofmt the file before
With both flags set to false, this CL causes no gofmt formatting change. We
can turn them on if we like it (and then remove the flags). Will submit with
flags disabled.
R=rsc, r
http://go/go-review/1025015
- enabled for function declarations (not just function literals)
- applied gofmt -w $GOROOT/src
(look for instance at src/pkg/debug/elf/elf.go)
R=r, rsc
CC=go-dev
http://go/go-review/1026006
both 64 and 32 bit platforms.
Also, uncomment some tests that were still commented out from
debugging.
R=gri, rsc
CC=go-dev
http://go/go-review/1026003
- this ensures better comment formatting in template.go and codec_test.go
- it affects only 2 files of all files that have been gofmt'ed already,
see separate CL (it fixes the same issue in those files)
R=rsc
http://go/go-review/1023002
Make RSA and X509 build by using big. (This involves commenting out
key generation for now since I haven't written Miller-Rabin in big
yet.)
Add entries to the Makefile.
R=rsc
CC=go-dev
http://go/go-review/1022005
(With hindsight, record_process might have been designed wrong, but it
works for now. It'll get redrawn when client support is added.)
R=rsc
CC=r
http://go/go-review/1018032
(the one-line struct types used in composite literals will become
one line again in another cleanup round; don't worry about them now)
R=r
http://go/go-review/1016056
- slightly better output in general
- extra indentation where we could do without
(however that seems better for now that not having the indentation
where it is needed)
- no information is lost, so a future better approach can fix these
places up again
R=rsc
http://go/go-review/1017050
- preserve (some) indentation of comment text for /*-style comments
even if the first comment line does not contain any text that might
suggest the "correct" indentation
- enabled because otherwise existing larger comments get re-formatted
(this will not introduce a lot of changes since comments of this
kind - until now - were not changed with respect to indentation)
R=rsc
http://go/go-review/1016047
picked up by gofmt (they should not be modified) or godoc
(testdata should not be considered a potential package
directory)
R=rsc
http://go/go-review/1018045
- at the moment this also affects single-line function declarations
because we have disabled them at the moment (but not single-line
function literals)
R=rsc
http://go/go-review/1016040
For testing it can be useful to use a null random source (one which
always returns zero) to remove non-determinism from the tests.
However, when performing RSA blinding, the random blind ends up being
zero and it's hard to reverse a multiplication by zero.
R=rsc
CC=go-dev
http://go/go-review/1018033
(which I uploaded at home and thus can't upload from here).
CL 1018027 was reviewed.
- added comments to scope.go
- commented out some code that is not yet needed
(and which showed up prominently in the documentation)
R=rsc
http://go/go-review/1017017
- start new sections if a field/method declaration spans multiple lines;
this avoids tabs from the previous line affecting the next field/method
R=rsc
http://go/go-review/1017015
- nodes.go implements ast node formatting
- printer.go implements the core printing functionality
and public interface
No code changes except for updating the import clauses
and adding a top-level comment to nodes.go.
R=rsc
http://go/go-review/1016026
added ReadFrom/WriteTo for packet protocols like UDP.
simplified the net.Conn interface.
added new net.PacketConn interface for packet protocols.
implemented proper UDP listener.
cleaned up LocalAddr/RemoteAddr methods - cache in netFD.
threw away various unused methods.
an interface change:
introduced net.Addr as a network address interface,
to avoid conversion of UDP host:port to string and
back for every ReadFrom/WriteTo sequence.
another interface change:
since signature of Listener.Accept was changing anyway,
dropped the middle return value, because it is available
as c.RemoteAddr(). (the Accept signature predates the
existence of that method.)
Dial and Listen still accept strings, but the proto-specific
versions DialTCP, ListenUDP, etc. take net.Addr instead.
because the generic Dial didn't change and because
no one calls Accept directly (only indirectly via the http
server), very little code will be affected by these interface
changes.
design comments welcome.
R=p
CC=go-dev, r
http://go/go-review/1018017
Firstly, with -Werror, GCC switched to printing warnings starting
with "error:". Widening the string matches solves this as the messages
are otherwise unchanged.
Secondly, GCC 4.4 outputs DWARF sections with with NUL bytes in all
the offsets and requires the relocation section for .debug_info to be
processed in order to result in valid DWARF data. Thus we add minimal
handling for relocation sections, which is sufficient for our needs.
BUG=1
Fixes#1.
R=rsc, iant
CC=go-dev
http://go/go-review/1017003
(previously not shown in godoc)
- simplify parsing of struct types (match code structure for parsing interface types)
R=rsc, r
http://go/go-review/1016019
* rename PORT.sh -> mkall.sh (hopefully more obvious),
change behavior: run commands by default.
* pull more constants out of #defines automatically,
instead of editing large lists by hand.
* add Recvfrom, Sendto
add os.O_EXCL.
R=r
http://go/go-review/1017009
