Rob Pike [Fri, 14 Sep 2012 22:25:37 +0000 (15:25 -0700)]
text/template: towards better errors
Give the right name for errors, and add a test to check we're
getting the errors we expect.
Also fix an ordering bug (calling add after stopParse) that
caused a nil indirection rather than a helpful error.
Fixes #3280.
Command.Start could crash before if no fds were available
because a nil *os.File of /dev/null was added to the cleanup
list, which crashed before returning the proper error.
Robert Griesemer [Fri, 14 Sep 2012 18:31:56 +0000 (11:31 -0700)]
spec: clarify section on string types
Strings happen to be represented similarly to
byte slices internally, but they don't quite
behave like them: While strings can be indexed,
sliced, and have their len() taken like byte
slices, string elements are not addressable,
make() and cap() is not supported, range loops
operate differently, and they are immutable (and
thus behave like values rather then references).
Fixes #4018.
R=r, rsc, iant, ken
CC=golang-dev
https://golang.org/cl/6503116
exp/locale/collate: changed API to allow access to different locales through New(),
instead of variables. Several reasons:
- Encourage users of the API to minimize the number of creations and reuse Collate objects.
- Don't rule out the possibility of using initialization code for collators. For some locales
it will be possible to have very compact representations that can be quickly expanded
into a proper table on demand.
Other changes:
- Change name of root* vars to main*, as the tables are shared between locales.
- Added Locales() method to get a list of supported locales.
Alex Brainman [Fri, 14 Sep 2012 02:53:30 +0000 (12:53 +1000)]
misc/dashboard/builder: use c:\ as default buildroot on windows
We have some tests (misc/cgo/test) that are disabled only because
they will fail to run on go builder - see issue 3358 for details.
This change will allow us to enable these tests.
Adam Langley [Thu, 13 Sep 2012 15:00:16 +0000 (11:00 -0400)]
crypto/tls: allow certificates and key to be in either order.
X509KeyPair wasn't really supposed to allow the certificate and
key to be in the same file, but it did work if you put the key
first. Since some HTTPS servers support loading keys and certs
like this, this change makes it work in either order.
crypto/rsa: reject PublicKey.E if it won't fit in a 32-bit int
Right now we only have 32-bit ints so that's a no-op.
Took the opportunity to check for some other invalid values too.
Suggestions for additions or modifications welcome.
This change messes with anchor links. It obscures the item being linked to.
I don't see a way around it. Undoing for now.
Fixes #4071.
««« original CL description
doc css: topbar sticks to the top of large windows.
Rationale: for large screens, the convenience of not having to scroll
to the top of the page to do a search outweighs having less vertical
space.
Tested with Chrome, Firefox, Safari with various window and text sizes.
go tool dist env -w is supposed to print a Windows batch file.
Normally Windows will execute batch files without \r before \n,
but issue 3060 reports that if the file ends up containing paths
written in Chinese, Windows 7 cannot execute it without the \r.
So add the \r.
This CL also adds support for marking the likelyness of IF nodes in the AST being true. This feature is being used here to mark the slow path as unlikely.
src/pkg/runtime:
benchmark old ns/op new ns/op delta
BenchmarkConvT2IUintptr 16 1 -91.63%
The key difference is that
0011 (main.go:5) MOVQ $4,BX
0012 (main.go:5) ADDQ CX,BX
0013 (main.go:5) MOVQ BX,CX
has changed to
0011 (main.go:5) ADDQ $4,CX
cmd/6g, cmd/8g: do not LEA[LQ] interfaces when calling methods.
It is enough to load directly the data word and the itab word
from memory, so we save a LEA instruction for each method call,
and allow elimination of some extra temporaries.
Brian Slesinskya [Tue, 11 Sep 2012 01:38:47 +0000 (11:38 +1000)]
doc css: topbar sticks to the top of large windows.
Rationale: for large screens, the convenience of not having to scroll
to the top of the page to do a search outweighs having less vertical
space.
Tested with Chrome, Firefox, Safari with various window and text sizes.
exp/locale/collate: added indices to builder for reusing blocks between locales.
Refactored build + buildTrie into build + buildOrdering.
Note that since the tailoring code is not checked in yet, all tailorings are identical
to root. The table therefore should not and does not grow at this point.
Rob Pike [Fri, 7 Sep 2012 17:28:24 +0000 (10:28 -0700)]
spec: an initial BOM can be ignored
After further deliberation, let's back down to the Unicode proposal.
Ignoring aBOMinations anywhere means that things like
grep unsafe *.go
might fail because there's a BOM in the middle: unBOMsafe.
Joel Sing [Fri, 7 Sep 2012 03:32:40 +0000 (13:32 +1000)]
cgo: use debug data section for ELF
When generating enums use the debug data section instead of the
DWARF debug info, if it is available in the ELF file. This allows
mkerrors.sh to work correctly on OpenBSD/386 and NetBSD/386.
Dave Cheney [Thu, 6 Sep 2012 07:50:26 +0000 (17:50 +1000)]
crypto/tls: fix data race on conn.err
Fixes #3862.
There were many areas where conn.err was being accessed
outside the mutex. This proposal moves the err value to
an embedded struct to make it more obvious when the error
value is being accessed.
As there are no Benchmark tests in this package I cannot
feel confident of the impact of this additional locking,
although most will be uncontended.
Nigel Tao [Thu, 6 Sep 2012 01:10:47 +0000 (11:10 +1000)]
image/jpeg: fix quantization tables to be in zig-zag order, not natural
order.
JPEG images generated prior to this CL are still valid JPEGs, as the
quantization tables used are encoded in the wire format. Such JPEGs just
don't use the recommended quantization tables.
The old code was a depth first graph traversal that could, under the
right conditions, end up re-exploring the same subgraphs multiple
times, once for each way to arrive at that subgraph at a given depth.
The new code uses a breadth first search to make sure that it only
visits each reachable embedded struct once.
Also add fast path for the trivial case.
benchmark old ns/op new ns/op delta
BenchmarkFieldByName1 1321 187 -85.84%
BenchmarkFieldByName2 6118 5186 -15.23%
BenchmarkFieldByName3 8218553 42112 -99.49%
R=gri, r
CC=golang-dev
https://golang.org/cl/6458090
Alan Donovan [Tue, 4 Sep 2012 18:40:49 +0000 (14:40 -0400)]
runtime: discard SIGPROF delivered to non-Go threads.
Signal handlers are global resources but many language
environments (Go, C++ at Google, etc) assume they have sole
ownership of a particular handler. Signal handlers in
mixed-language applications must therefore be robust against
unexpected delivery of certain signals, such as SIGPROF.
The default Go signal handler runtime·sigtramp assumes that it
will never be called on a non-Go thread, but this assumption
is violated by when linking in C++ code that spawns threads.
Specifically, the handler asserts the thread has an associated
"m" (Go scheduler).
This CL is a very simple workaround: discard SIGPROF delivered to non-Go threads. runtime.badsignal(int32) now receives the signal number; if it returns without panicking (e.g. sig==SIGPROF) the signal is discarded.
I don't think there is any really satisfactory solution to the
problem of signal-based profiling in a mixed-language
application. It's not only the issue of handler clobbering,
but also that a C++ SIGPROF handler called in a Go thread
can't unwind the Go stack (and vice versa). The best we can
hope for is not crashing.
Note:
- I've ported this to all POSIX platforms, except ARM-linux which already ignores unexpected signals on m-less threads.
- I've avoided tail-calling runtime.badsignal because AFAICT the 6a/6l don't support it.
- I've avoided hoisting 'push sig' (common to both function calls) because it makes the code harder to read.
- Fixed an (apparently incorrect?) docstring.
Patrick Crosby [Tue, 4 Sep 2012 01:27:20 +0000 (11:27 +1000)]
net/http/pprof: updated documentation (run an http server)
Added instructions for starting an http server
to the godoc header for this package. With the old
instructions, the example "go tool pprof..." commands
wouldn't work unless you happen to be running an http
server on port 6060 in your application.
Shenghou Ma [Sun, 2 Sep 2012 19:49:03 +0000 (03:49 +0800)]
doc/progs: use test/run.go for testing on Windows
cgo[1-4].go, go1.go couldn't be tested now
(cgo[1-4].go can only be tested when cgo is enabled, go1.go
contain a list of filenames in the current directory)
codereview.py: correct error handling without decorator
The decorator hides the number of function arguments from Mercurial,
so Mercurial cannot give proper error messages about commands
invoked with the wrong number of arguments.
Left a 'dummy' hgcommand decorator in place as a way to document
what functions are hg commands, and just in case we need some other
kind of hack in the future.
There was mail on golang-nuts a few weeks ago
from someone who understood the message perfectly
and knew he had a cyclic dependency but assumed
that Go, like Python or Java, was supposed to handle it.
R=golang-dev, bradfitz, dave
CC=golang-dev
https://golang.org/cl/6488069