Robert Griesemer [Thu, 6 Dec 2012 17:17:20 +0000 (09:17 -0800)]
spec: type assertions and type switches must be valid
The spec didn't preclude invalid type assertions and
type switches, i.e., cases where a concrete type doesn't
implement the interface type in the assertion in the first
place. Both, the gc and gccgo compiler exclude these cases.
This is documenting the status quo.
Also:
- minor clean up of respective examples
- added sentence about default case in select statements
Fixes #4472.
R=rsc, iant, r, ken
CC=golang-dev
https://golang.org/cl/6869050
Rémy Oudompheng [Thu, 6 Dec 2012 06:20:03 +0000 (07:20 +0100)]
cmd/6c, cmd/8c: add fixjmp step to regopt.
The fixjmp step eliminates redundant chains of JMP
instructions that are produced by the compiler during
code generation.
It is already implemented in gc, and can be adapted to 6c/8c with
the exception that JMPs refer to destination by pc instead of by
pointer. The algorithm is modified to operate on Regs instead of Progs
for this reason. The pcs are already restored later by regopt.
Russ Cox [Thu, 6 Dec 2012 05:00:20 +0000 (00:00 -0500)]
cmd/ld: skip 0-length write in cwrite
The 0-length part is fine, but some callers that write 0 bytes
also pass nil as the data pointer, and the Plan 9 kernel kills the
process with 'invalid address in sys call' in that case.
Brad Fitzpatrick [Thu, 6 Dec 2012 03:08:42 +0000 (19:08 -0800)]
net/http: fix bug parsing http_proxy lacking a protocol
Per the curl man page, the http_proxy configuration can be
of the form:
[protocol://]<host>[:port]
And we had a test that <ip>:<port> worked, but if
the host began with a letter, url.Parse parsed the hostname
as the scheme instead, confusing ProxyFromEnvironment.
Dave Cheney [Wed, 5 Dec 2012 21:01:33 +0000 (08:01 +1100)]
cmd/5g: use MOVB for fixed array nil check
Fixes #4396.
For fixed arrays larger than the unmapped page, agenr would general a nil check by loading the first word of the array. However there is no requirement for the first element of a byte array to be word aligned, so this check causes a trap on ARMv5 hardware (ARMv6 since relaxed that restriction, but it probably still comes at a cost).
Switching the check to MOVB ensures alignment is not an issue. This check is only invoked in a few places in the code where large fixed arrays are embedded into structs, compress/lzw is the biggest offender, and switching to MOVB has no observable performance penalty.
Thanks to Rémy and Daniel Morsing for helping me debug this on IRC last night.
Ian Lance Taylor [Wed, 5 Dec 2012 00:51:51 +0000 (16:51 -0800)]
test: tweak bug273.go to not have constant len < cap when calling make
The current spec says that when calling make, if both len and
cap are constant, it is an error if len > cap. The gc
compiler does not yet implement that, but when it does this
will need to change. Changing it now for the benefit of
gccgo.
Robert Griesemer [Mon, 3 Dec 2012 22:23:41 +0000 (14:23 -0800)]
spec: channel operations are restricted by the channel direction
Also:
- 'for' statements with a range clause do not accept send-only
channels
- '_, _ = range ch' is not equivalent to "_ = range ch" if ch
is a channel (rewriting the latter to the former leads to
an invalid range clause).
These clarifications document the status quo.
R=rsc, r, iant, ken
CC=golang-dev
https://golang.org/cl/6874053
Dave Cheney [Mon, 3 Dec 2012 21:27:30 +0000 (08:27 +1100)]
cmd/dist: support building on debian/kFreeBSD
The debian/kFreeBSD project uses the FreeBSD kernel and the debian userspace. From our point of view, this is freebsd not linux as GOOS talks about the kernel syscall interface, not the userspace (although cgo alters that). As debian/kFreeBSD is experimental at this time, I do not think it is worth the effort of duplicating all the freebsd specific code so this is proposal represents a reasonable workaround.
Currently cgo is not supported, make.bash will detect this and disable cgo automatically during the build.
dfc@debian:~/go/src$ uname -a
GNU/kFreeBSD debian 8.1-1-686 #0 Sat Jul 21 17:02:04 UTC 2012 i686 i386 Intel(R) Core(TM) i5-2415M CPU @ 2.30GHz GNU/kFreeBSD
dfc@debian:~/go/src$ ../bin/go version
go version devel +d05272f402ec Sat Dec 01 15:15:14 2012 -0800
Johan Euphrosine [Sat, 1 Dec 2012 22:11:46 +0000 (14:11 -0800)]
math/rand: remove noop iteration in Perm
The first iteration always do `m[0], m[0] = m[0], m[0]`, because
`rand.Intn(1)` is 0.
fun note: IIRC in TAOCP version of this algorithm, `i` goes
backward (n-1->1), meaning that the "already" shuffled part of the
array is never altered betweens iterations, while in the current
implementation the "not-yet" shuffled part of the array is
conserved between iterations.
Dmitriy Vyukov [Fri, 30 Nov 2012 06:29:41 +0000 (10:29 +0400)]
runtime: better stack traces in race reports
When a race happens inside of runtime (chan, slice, etc),
currently reports contain only user file:line.
If the line contains a complex expression,
it's difficult to figure out where the race exactly.
This change adds one more top frame with exact
runtime function (e.g. runtime.chansend, runtime.mapaccess).
Dmitriy Vyukov [Fri, 30 Nov 2012 06:27:43 +0000 (10:27 +0400)]
cmd/gc: racewalk: collect stack traces in sync and sync/atomic
W/o this change stack traces do not show from where sync.Once()
or atomic.XXX was called.
This change add funcenter/exit instrumentation to sync/sync.atomic
packages.
Brad Fitzpatrick [Fri, 30 Nov 2012 02:00:51 +0000 (18:00 -0800)]
net/http: don't send chunked encoding on 204 responses
RFC 2616: "The 204 response MUST NOT include a message-body,
and thus is always terminated by the first empty line after
the header fields."
Previously we'd trigger chunked encoding by default on
responses, and then when finishing the request we'd write the
chunk trailers, which counted as a message-body.
Test creates 2 tcp connections for put and get. Make sure
these are closed properly after test is over, otherwise
server hangs waiting for connection to be closed.
R=golang-dev, bradfitz, dave
CC=golang-dev
https://golang.org/cl/6842109
Dave Cheney [Wed, 28 Nov 2012 00:29:25 +0000 (11:29 +1100)]
net: move deadline logic into pollServer
Update #4434.
The proposal attempts to reduce the number of places where fd,{r,w}deadline is checked and updated in preparation for issue 4434. In doing so the deadline logic is simplified by letting the pollster return errTimeout from netFD.Wait{Read,Write} as part of the wakeup logic.
The behaviour of setting n = 0 has been restored to match rev 2a55e349097f, which was the previous change to fd_unix.go before CL 6851096.
Robert Griesemer [Tue, 27 Nov 2012 18:29:49 +0000 (10:29 -0800)]
go/format: Package format implements standard formatting of Go source.
Package format is a utility package that takes care of
parsing, sorting of imports, and formatting of .go source
using the canonical gofmt formatting parameters.
Use go/format in various clients instead of the lower-level components.
Jan Ziak [Tue, 27 Nov 2012 18:04:59 +0000 (13:04 -0500)]
runtime: use reflect·call() to enter the function gc()
Garbage collection code (to be merged later) is calling functions
which have many local variables. This increases the probability that
the stack capacity won't be big enough to hold the local variables.
So, start gc() on a bigger stack to eliminate a potentially large number
of calls to runtime·morestack().
Rick Arnold [Tue, 27 Nov 2012 17:58:27 +0000 (12:58 -0500)]
regexp: add Split
As discussed in issue 2672 and on golang-nuts, this CL adds a Split() method
to regexp. It is based on returning the "opposite" of FindAllString() so that
the returned substrings are everything not matched by the expression.
John Graham-Cumming [Tue, 27 Nov 2012 15:21:43 +0000 (10:21 -0500)]
log/syslog: correct message format
The syslog implementation was not correctly implementing the
traditional syslog format because it had a confused notion of
'priority'. syslog priority is not a single number but is, in
fact, the combination of a facility number and a severity. The
previous Go syslog implementation had a single Priority that
appeared to be the syslog severity and no way of setting the
facility. That meant that all syslog messages from Go
programs appeared to have a facility of 0 (LOG_KERN) which
meant they all appeared to come from the kernel.
Also, the 'prefix' was in fact the syslog tag (changed the
internal name for clarity as the term tag is more widely used)
and the timestamp and hostname values were missing from
messages.
With this change syslog messages are generated in the correct
format with facility and severity combined into a priority,
the timestamp in RFC3339 format, the hostname, the tag (with
the PID in [] appened) and the message.
The format is now:
<PRI>1 TIMESTAMP HOSTNAME TAG[PID]: MSG
The TIMESTAMP, HOSTNAME and PID fields are filled in
automatically by the package. The TAG and the MSG are supplied
by the user. This is what rsyslogd calls TraditionalFormat and
should be compatible with multiple systems.
Dmitriy Vyukov [Tue, 27 Nov 2012 08:18:54 +0000 (12:18 +0400)]
net: fix flaky test
The test failed on one of the builders with:
timeout_test.go:594: ln.Accept: accept tcp 127.0.0.1:19373: use of closed network connection
http://build.golang.org/log/e83f4a152b37071b9d079096e15913811ad296b5
Brad Fitzpatrick [Mon, 26 Nov 2012 21:31:02 +0000 (13:31 -0800)]
net/http: fix Transport races & deadlocks
Thanks to Dustin Sallings for exposing the most frustrating
bug ever, and for providing repro cases (which formed the
basis of the new tests in this CL), and to Dave Cheney and
Dmitry Vyukov for help debugging and fixing.
This CL depends on submited pollster CLs ffd1e075c260 (Unix)
and 14b544194509 (Windows), as well as unsubmitted 6852085.
Some operating systems (OpenBSD, NetBSD, ?) may still require
more pollster work, fixing races (Issue 4434 and
http://goo.gl/JXB6W).
Tested on linux-amd64 and darwin-amd64, both with GOMAXPROCS 1
and 4 (all combinations of which previously failed differently)
Fixes #4191
Update #4434 (related fallout from this bug)
Daniel Morsing [Mon, 26 Nov 2012 21:23:13 +0000 (22:23 +0100)]
cmd/gc: Make sure bools lose idealness when used with logical operators.
Bools from comparisons can be assigned to all bool types, but this idealness would propagate through logical operators when the result should have been lowered to a non-ideal form.
Robert Griesemer [Mon, 26 Nov 2012 21:14:04 +0000 (13:14 -0800)]
go/printer: Permit declaration and statement lists as input.
Also: Can set base indentation in printer.Config: all code
is going to be indented by at least that amount (except for
raw string literals spanning multiple lines, since their
values must not change).
Rémy Oudompheng [Mon, 26 Nov 2012 20:51:48 +0000 (21:51 +0100)]
cmd/6l, cmd/8l: emit no-ops to separate zero-stack funcs from nosplits.
The stack overflow checker in the linker uses the spadj field
to determine whether stack space will be large enough or not.
When spadj=0, the checker treats the function as a nosplit
and emits an error although the program is correct.
Robert Griesemer [Mon, 26 Nov 2012 20:49:04 +0000 (12:49 -0800)]
exp/types: fixed field/method lookup
also:
- composite literal checking close to complete
- cleaned up parameter, method, field checking
- don't let panics escape type checker
- more TODOs eliminated