The gccgo compiler does not provide a gc-style GOROOT with standard
library sources. The effect is that guru may not fully work when using
gccgo. However, it can fully work if the GOROOT environment variable
points to valid gc-style GOROOT. In that case, make it work by telling
the go/build package to use gc mode.
Change-Id: Iadff8be61be8cc9a7ff2ca0a067b116b62895451
Reviewed-on: https://go-review.googlesource.com/117997
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
This CL adds a missing closing double-quote to the documentation string
of gomvpkg.
Change-Id: I000fe33fb9ddc5342bbf51c36595a884d6aa4eab
Reviewed-on: https://go-review.googlesource.com/114115
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
- Setting the GOROOT build path to the value passed from the command line.
- Clarified the return values to named parameters for extra clarity.
- And while here, added some missed out error handling. Just logging the error
to preserve original behavior.
Fixesgolang/go#13296
Change-Id: I91427eee790928a3cfb51ae207747e9a17bd5496
Reviewed-on: https://go-review.googlesource.com/110275
Run-TryBot: Andrew Bonventre <andybons@golang.org>
Reviewed-by: Andrew Bonventre <andybons@golang.org>
I felt guilty about leaving 2% on the table in CL 108878,
so I thought I'd get it a different way.
Teach readFile to accept a re-usable bytes.Buffer to read into,
to reduce the amount of garbage created.
To limit the possible memory impact of giant files,
only re-use the buffer for the duration of a single package.
Even that is enough to help.
name old time/op new time/op delta
Referrers 4.67s ± 3% 4.58s ± 2% -1.96% (p=0.029 n=10+10)
name old user-time/op new user-time/op delta
Referrers 16.5s ± 3% 15.8s ± 1% -4.39% (p=0.000 n=10+8)
name old sys-time/op new sys-time/op delta
Referrers 16.1s ± 3% 15.9s ± 3% ~ (p=0.218 n=10+10)
This work supported by Sourcegraph.
Change-Id: I594ef25c0fd5ccb766ff5b98dbbd1a75a7a4f957
Reviewed-on: https://go-review.googlesource.com/108935
Reviewed-by: Alan Donovan <adonovan@google.com>
This simplifies and unifies the code.
It also improves the output:
Now the initial result is referred to from the perspective of
the query site, rather than from the perspective of the package
containing the query object, which seems more natural.
This work supported by Sourcegraph.
Change-Id: Ie7a77d39dd6568334e72d894dc66d35494ed4ac4
Reviewed-on: https://go-review.googlesource.com/108936
Reviewed-by: Alan Donovan <adonovan@google.com>
This change parallelizes the outer loop in globalReferrersPkgLevel,
which loops over packages to inspect.
There is also an easily parallelizable inner loop.
However, parallelizing it adds complication
(deffiles needs a mutex, inQueryPackage requires a wait group)
and offers only a 2% speed-up.
Benchmarks for this change, looking for encoding/json.MarshalIndent:
name old time/op new time/op delta
Referrers 5.31s ± 2% 4.67s ± 3% -11.95% (p=0.000 n=10+10)
name old user-time/op new user-time/op delta
Referrers 15.9s ± 2% 16.5s ± 3% +3.71% (p=0.000 n=10+10)
name old sys-time/op new sys-time/op delta
Referrers 15.7s ± 3% 16.1s ± 3% +2.73% (p=0.011 n=10+10)
Fixesgolang/go#24272
Updates golang/go#25017
This work supported by Sourcegraph.
Change-Id: I5dcda9017103cdff59d0ffdf5e87d2c2c955a33a
Reviewed-on: https://go-review.googlesource.com/108878
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
Global, package-level queries can be done without typechecking.
This makes it significantly cheaper.
Instead of typechecking, consider all packages that directly
import the query package. In each of those packages, filter out:
* files that don't contain the query name
* files that don't import the query package
Fully parse the remaining files and look for selector expressions
matching the query package and name.
There are a few twists (dot imports, the query package, xtests).
They are described, along with how they are handled,
in the large comment at the beginning of globalReferrersPkgLevel.
On my machine and GOPATH, this reduces typical time
needed to find referrers to encoding/json.MarshalIndent
from (roughly)
real 0m39.946s
user 2m27.844s
sys 0m54.774s
to (roughly)
real 0m5.687s
user 0m15.793s
sys 0m16.001s
The processing of packages could be parallelized;
that is left for a future change, to ease reviewing.
Benefits from that will be limited;
building the reverse import graph accounts for 73%
of the runtime, and that is dominated by syscalls.
Optimization idea from Alan Donovan.
This work supported by Sourcegraph.
Change-Id: Ib19e25fcdcb27673fb03d7300dba2a53198901ad
Reviewed-on: https://go-review.googlesource.com/97800
Reviewed-by: Alan Donovan <adonovan@google.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Previously, goyacc's line number tracking would run off by one whenever
a multiline comment was used inside of an action, like so:
expr:
TOKEN
{
/* Hello. */
}
This is because the character after the multi-line comment close marker
(i.e., the character immediately after `*/`) was blindly printed out
instead of properly accounted for. A newline character after a
multi-line comment would fail to increment lineno, for example, and any
error messages generated after that point would refer to the wrong line.
Similarly, a variable reference after a multi-line comment, like
$$.val = &someStruct{Value: /* oops */$1}
would copy the $ literally into the resulting Go code. (This was not a
problem in practice because multi-line comments are typically followed
by whitespace.)
Adjust the control flow so the character after the multi-line comment
close marker character gets run through the relevant switch statement
and accounted for.
Change-Id: I276b6ffdb7626101f76811b7ee4804bacc6ef740
Reviewed-on: https://go-review.googlesource.com/110495
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
If test files are not included here,
then results in those files will be missing
for queries within a single package.
This work supported by Sourcegraph.
Change-Id: I8136ea250ba980179228d9b408d9e7409c49a873
Reviewed-on: https://go-review.googlesource.com/97799
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
This make it possible to easily run individual guru tests.
It also slightly speeds up the overall test run,
from 23s to 18s on my machine;
the TestGuru/calls is now the limiting factor,
by a significant margin.
This work supported by Sourcegraph.
Change-Id: If61ebf1cc60441a65274f3fddd31f69c7ca23b48
Reviewed-on: https://go-review.googlesource.com/108876
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
Currently, links inside blog articles are absolute links to golang.org.
But when a godoc server is run locally, the blog package should
serve local links pointing to the local godoc server. It is not possible
to simply change the links inside the blog templates to relative urls
because the blog articles are independant pages on their own.
And moreover, they are served from blog.golang.org.
Rather, the blog package consumes and serves blog articles.
So, a flag was added in the Config struct to denote whether
to convert the links or not. This flag is then set from the
call site in godoc package where the blog server is initialized from.
This was required because "golang.org/x/tools/blog" is a package
which can be used by other code to serve blog pages and not just godoc.
This preserves existing functionality for all working code which
imports "golang.org/x/tools/blog" and changes the functionality only
when a godoc server is run locally.
And while here, replace relevant bytes.Buffer occurences
with strings.Builder.
Fixesgolang/go#22681
Change-Id: I7dbf9c5f2f93fd0b7e17915238de1c084fcd1431
Reviewed-on: https://go-review.googlesource.com/105835
Reviewed-by: Andrew Bonventre <andybons@golang.org>
Run-TryBot: Andrew Bonventre <andybons@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Separate the templates of package root directory and other
directories. This removes several if-else conditions and
makes the template code much cleaner.
To be merged after CL 95835.
For golang/go#3305
Change-Id: I435441dbe214e4c89271483173df7c30a42ee28f
Reviewed-on: https://go-review.googlesource.com/101295
Run-TryBot: Andrew Bonventre <andybons@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Andrew Bonventre <andybons@golang.org>
They exist and are considered to be worth listing based on the decision
made in golang/go#24432. They weren't listed previously. This change
fixes that.
Document the remaining subrepos that are not meant to be listed (per
decision in golang/go#24432), so it's clear that it's intentional for
them to not be visible at https://golang.org/pkg/#subrepo.
Closesgolang/go#24432.
Updates golang/go#14304.
Change-Id: Icc50ebfcdbc490c32519d92a1a838eb3f54c823d
Reviewed-on: https://go-review.googlesource.com/103075
Reviewed-by: Austin Clements <austin@google.com>
The godoc in the test was indexing sources in the default GOPATH.
If the default GOPATH pointed to local workspace, test would timeout.
The fix is to supply GOPATH set to non-existing path.
Fixesgolang/go#24504
Change-Id: Iedf044cdec78d5c5642105650ad8ec17aa10a5ba
Reviewed-on: https://go-review.googlesource.com/102295
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This flag includes unexported identifiers in command-line mode.
It is equivalent to ?m=all in web mode.
Fixesgolang/go#8093
Change-Id: I1e5a69626929d3430638d900f3e975b272a98c90
Reviewed-on: https://go-review.googlesource.com/99435
Run-TryBot: Andrew Bonventre <andybons@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Andrew Bonventre <andybons@golang.org>
This regression is due to https://go-review.googlesource.com/c/go/+/100235
which made the go/scanner behavior match the compiler's scanner by
not modifying filenames in line directives in any form. Specifically,
relative filenames now remain relative. This makes it easier to reason
about the behavior of the scanner but may require clients to adjust
those filenames.
If removing the longer path in the .golden file is not satisfactory,
the (scanner) client may have to massage positions returned via line
directives; or better, the line directives are created with extended
path information.
R=adonovan
Change-Id: I836571a6dafef5e6d81e4c7add203c994e107055
Reviewed-on: https://go-review.googlesource.com/101015
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Andrew Bonventre <andybons@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
In cmd/goimports, allow for the -local flag to accept a comma-separated
list of import path prefixes. Also, update the imports package
accordingly to support this.
Fixesgolang/go#19188
Change-Id: I083d584df8c3a77532f0f66e9c5d970960180e0d
Reviewed-on: https://go-review.googlesource.com/85397
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
In cmd/present, a mode was being passed to the function parse, but it
wasn't actually being used. Use it.
In go/ssa, checkFinalInstr received an idx integer but it doesn't
actually need it. Get rid of it.
Lastly, in imports, findImportStdlib always returned rename==false. Get
rid of that result parameter.
Change-Id: I719006b69ee80a3ef4b0ea24c1c206016a7e304b
Reviewed-on: https://go-review.googlesource.com/93596
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Currently, in http mode the server blocks until the corpus
has been initialized. This can cause considerable delay
if the user workspace is significantly large and the files
are not present in the buffer cache.
This CL spawns off the initialization in a separate goroutine
if httpMode is set and turns on a flag when it's done.
The http handler checks the flag and returns an error response
if it has not been set.
The check is only performed for the path prefixes handled by the
handlerServer struct. Other paths do not call the GetPageInfo() function
and hence can return immediately. This preserves maximum responsiveness
of the server.
Also adds an additional print statement in verbose mode
Note: This is a re-do of a previous CL golang.org/cl/88695 which was
incorrect committed without running tests. This CL fixes that test.
Fixesgolang/go#13278
Change-Id: I80c801f32af007312090d3783a2ea2c6f92cad66
Reviewed-on: https://go-review.googlesource.com/93215
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Currently, in http mode the server blocks until the corpus
has been initialized. This can cause considerable delay
if the user workspace is significantly large and the files
are not present in the buffer cache.
This CL spawns off the initialization in a separate goroutine
if httpMode is set and turns on a flag when it's done.
The http handler checks the flag and returns an error response
if it has not been set.
The check is only performed for the path prefixes handled by the
handlerServer struct. Other paths do not call the GetPageInfo() function
and hence can return immediately. This preserves maximum responsiveness
of the server.
Also adds an additional print statement in verbose mode
Fixesgolang/go#13278
Change-Id: I0505acc1c190423d09fb199b11ca86e0400e84d4
Reviewed-on: https://go-review.googlesource.com/88695
Reviewed-by: Andrew Bonventre <andybons@golang.org>
Now that play.golang.org and sandbox[-flex].golang.org have been
merged, proxy requests to the former.
Change-Id: I7d18d0494fd54c2357dc53952fa458ceb1380aca
Reviewed-on: https://go-review.googlesource.com/86253
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
This can be very helpful if you lay out each value's string
representation like this:
and // &
andAnd // &&
or // |
orOr // ||
Without the use of comments, it's impossible to use stringer with these
names as the characters & and | cannot form valid identifiers in a Go
program.
Fixes#20483.
Change-Id: I4d36c74059dd48ae3a5e09b70a429a75853ef179
Reviewed-on: https://go-review.googlesource.com/44076
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Rob Pike <r@golang.org>
getgo doesn't work on plan9. Skip it entirely.
And skip the massive slow godoc start-up test. Not worth it.
Change-Id: If062b7c4c8c7c5084e607ed22085657054c10ba9
Reviewed-on: https://go-review.googlesource.com/80737
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
In the generated code, we want to pull in as few dependencies as
possible. fmt is heavier than strconv, and the latter can be used with a
bit of extra code for the same results.
More importantly, this will allow stringer to be used in some std
packages that cannot import fmt, such as regexp/syntax. While strconv
lies in L2 in deps_test.go, fmt lies in L4.
This means that many other packages will also be able to use stringer
where it could be useful, such as path/filepath, os/exec, or io/ioutil.
Since some of these types may be 64-bit integers, use FormatInt instead
of Itoa to avoid overflows with int on 32-bit.
Also double-checked that the generated code is still formatted properly.
Change-Id: Iffb3bd2df5c94407705689719240aca0c7474a89
Reviewed-on: https://go-review.googlesource.com/77473
Reviewed-by: Ian Lance Taylor <iant@golang.org>
To trim a string prefix from the names when generating their final
strings. Add a simple test too.
There is no automatic detection of prefixes for now. That can be added
later, building on top of this first simple implementation.
Fixes#16539.
Change-Id: Ica37273ac74bb0a6cbd43e61823786963d86a492
Reviewed-on: https://go-review.googlesource.com/76650
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
When toolstash -cmp found diff, it does a second compilation
with extra flags -v -m=2, which are imcompatible with the
concurrent backend. Pass -c=1 in the second compilation.
Change-Id: I3c77069936da1829b68375a4a6c7f9bbe364247c
Reviewed-on: https://go-review.googlesource.com/60390
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
On mobile and tablets, it was very difficult to view slides because the
slides were not designed to be smaller than 1250x750.
This adds a function to the JS that uses CSS scaling to make the slides
fit on smaller displays.
Fixesgolang/go#21643
Change-Id: I68e9e2c1274aaf6396bf01d19ca023cddf76e2ec
Reviewed-on: https://go-review.googlesource.com/60270
Reviewed-by: Andrew Bonventre <andybons@golang.org>
Reviewed-by: Francesc Campoy Flores <campoy@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
Run-TryBot: Francesc Campoy Flores <campoy@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Caller is *ssa.Function. Its Pkg field is *ssa.Package. But ssa.Package
struct doesn't contain either field or method named Object.
So one gets the following error:
$ callgraph -format '{{.Caller.Pkg.Object.Path}} -> {{.Callee.Pkg.Object.Path}}' \
$(go env GOROOT)/src/net/http/triv.go | sort | uniq
callgraph: template: -format:1:9: executing "-format" at <.Caller.Pkg.Object.P...>: can't evaluate field Object in type *ssa.Package
ssa.Package contains Pkg field:
Pkg *types.Package // the corresponding go/types.Package
types.Package contains Path() method that returns the package path.
That appears to be the right way to get the "import path of the
enclosing package" with the current x/tools/go/ssa API.
Fixesgolang/go#20459.
Change-Id: I3cb7b61b767148f5c771d57f8f5e7e6ab20aea87
Reviewed-on: https://go-review.googlesource.com/43812
Reviewed-by: Avelino <t@avelino.xxx>
Reviewed-by: Alan Donovan <adonovan@google.com>
https://golang.org/cl/52390 was submitted too early with failing trybots.
This fixes it, hiding the cloud.google.com stuff behind a build tag,
used by the Dockerfile but not the Go build system.
Change-Id: I66c6b40d4b06bf6c763f3ab221c7997856bfc910
Reviewed-on: https://go-review.googlesource.com/52470
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jessie Frazelle <me@jessfraz.com>
Fixes expired cert issue too, either by virtue of newer autocert, or
that we're no longer hitting some LetsEncrypt rate limit.
But we have a cache now at least, so we should be nicer to LetsEncrypt
when the process dies. I did see some process crashes (old?) in the
logs. So maybe we were crashing and re-hitting LE often.
Verified that the cert now expires:
Sunday, October 29, 2017 at 9:57:00 PM Pacific
and that it's cached in GCS.
Fixesgolang/go#21251
Change-Id: I3def551ae8eef5df0ec7c51927c4d4bb52cd7a6a
Reviewed-on: https://go-review.googlesource.com/52390
Reviewed-by: Chris Broadfoot <cbro@golang.org>
Reviewed-by: Jessie Frazelle <me@jessfraz.com>
Run-TryBot: Jessie Frazelle <me@jessfraz.com>
There is nothing else we can do other than just showing a
message that user need to remove it from their PATH not
to conflict with the current installation.
Fixes#21217.
Change-Id: Ie65385f4d536d5bb789387ba0229f54f2ee793f0
Reviewed-on: https://go-review.googlesource.com/51930
Run-TryBot: Jaana Burcu Dogan <jbd@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Kevin Burke <kev@inburke.com>
