The -json and -c=N flags, formerly belonging only to the
go/packages-based {single,multi}checkers, are now supported by
unitchecker as well.
The no-op -source, -v, -all, and -tags flags, formerly belonging only
to unitchecker, have moved to the analysisflags package, which is
common to all checkers.
The -flags flag now reports all registered flags (except the
{single,multi}checker-only debugging flags) rather than just those
related to analyzers, allowing one to say: 'go vet -json' or 'go vet -c=1'.
The code for printing diagnostics, either plain or in JSON, has been
factored and moved into the common analysisflags package.
This CL depends on https://go-review.googlesource.com/c/go/+/149960 to
cmd/go, which causes 'go vet' to populate the ID field of the *.cfg.
This field is used as a key in the JSON tree.
Added basic tests of the new -json and -c unitchecker flags.
Change-Id: Ia7a3a9adc86de067de060732d2c200c58be3945a
Reviewed-on: https://go-review.googlesource.com/c/150038
Reviewed-by: Michael Matloob <matloob@golang.org>
(By "vet lite", we mean static tools that must be invoked by a build
system, such as 'go vet'.)
This CL publishes the former internal/unitchecker package.
Its misnamed Main function is renamed Run, and it has a new Main
that does the steps of a real main (log, flag, etc).
The motivation for this change is to reduce cmd/vet-lite to the point
of triviality to simplify the maintenance of the vendored copy of
x/tools in GOROOT, because GOROOT/src/cmd/vet will need a copy of that
logic. It is now essentially a one-liner.
Also, improve usage messages; analysisflags.PrintUsage wasn't
appropriate for all callers so it has been eliminated.
Each of {single,multi,unit}checker prints its own 1-line usage message.
Change-Id: I214c0e4ae7a2923eee8df3f7548341f2320cad2b
Reviewed-on: https://go-review.googlesource.com/c/149742
Reviewed-by: Michael Matloob <matloob@golang.org>
This change causes singlechecker and multichecker to exit with the
correct error code: 0 for success, 1 for load/analysis and other
errors, 3 for diagnostics. (We avoid 2 because the flag package uses
it.)
In JSON mode, errors in package loading, parsing, typechecking and
analysis are successfully in JSON format, with exit code 0.
+ Test.
Change-Id: Iaf130ed3d4cb3e747a628af6da8dc97d065aa869
Reviewed-on: https://go-review.googlesource.com/c/149603
Run-TryBot: Alan Donovan <adonovan@google.com>
Reviewed-by: Michael Matloob <matloob@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Also:
- add cmd/vet-lite, a version of cmd/vet that doesn't depend on
go/packages and must be run under "go vet". This will be vendored
into $GOROOT/src/cmd/vet.
- add an integration test for a unitchecker-based command under "go vet".
Change-Id: Id613dac2812816c6d6372fa6d1536c8d4e4c2676
Reviewed-on: https://go-review.googlesource.com/c/143418
Reviewed-by: Ian Cottrell <iancottrell@google.com>
Reviewed-by: Michael Matloob <matloob@golang.org>
The analysistest.Run function now applies a single analysis to
a set of packages, not just one, as this is necessary for testing
the "tests" Analyzer. The Run function also returns a richer
Result for each package, allowing a test to perform additional
checks if necessary.
I really don't understand how Gerrit decides whether to render
a file such as passes/tests/tests.go as a mv+edit or an add;
small changes to the CL seem to perturb the heuristic.
When reviewing these CLs please inspect the logical diff of
passes/vet/tests.go -> passes/tests/tests.go
Change-Id: I7812837278b20c8608ccbb6c709c675588a84db1
Reviewed-on: https://go-review.googlesource.com/c/140457
Reviewed-by: Michael Matloob <matloob@golang.org>
Run-TryBot: Michael Matloob <matloob@golang.org>
The ctrlflow Analyzer builds a control-flow graph (see
golang.org/x/tools/go/cfg) for each named and unnamed function in the
package.
It computes for each function whether it can never return, either
because the function is an intrinsic that stops the thread (e.g.
os.Exit), or because control never reaches a return statement, or
because the function inevitably calls another function that never
returns. For each such function it exports a noReturn fact.
This change also:
- adds 'inspect', another Analyzer that builds an optimized AST
traversal table for use by nearly every other Analyzer.
- changes analysistest.Run to return the analysis result to enable
further testing.
(This required changing it to analyze one package at a time,
which is no less efficient, and is the typical case.)
Change-Id: I877e2b2363a365a9976aa9c2719ad3fba4df2634
Reviewed-on: https://go-review.googlesource.com/c/139478
Reviewed-by: Michael Matloob <matloob@golang.org>
internal/checker
- don't display "[name.category]" in diagnostic
output. Most users don't care which analyzer reported the message.
The -json output retains this information.
- print better log messages for analyze -debug=f.
- print (non-JSON) text output to standard error, like a compiler would.
passes/pkgfact
- fix a nil deref panic when encountering non-renaming imports.
- require names to have underscores before and after (_x_)
as this avoids a huge number of spurious matches in (e.g.) the
syscall package.
- don't export empty facts.
Change-Id: I86c003b96521334e371f9d5fcea1323cd779d7f0
Reviewed-on: https://go-review.googlesource.com/c/139657
Reviewed-by: Michael Matloob <matloob@golang.org>
Run-TryBot: Michael Matloob <matloob@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This change adds support for testing the facts produced by an Analyzer
using a similar mechanism to the way it checks for diagnostics.
A "// want ..." comment may now contain a mixture of expectations for
diagnostics and facts. Diagnostics are indicated by a string literal,
as before. Facts are indicated by name:"regexp" where name identifies
the object (declared on the same line) with which the fact is
associated.
func neverReturns() { // want neverReturns:"noReturn"
for {}
}
Also:
- analysistest: report errors during package loading.
(We don't yet have a way to test RunDespiteErrors Analyzers in the
face of errors.)
- tests for Facts produced by findcall and pkgfacts.
(Findcall now produces facts just for testing.)
- Add String method to various Fact types.
Should the Fact interface have this method?
Change-Id: Ifa15fbd49d6ec3042b5fe9d3ebf22f4bdfdc8769
Reviewed-on: https://go-review.googlesource.com/139157
Reviewed-by: Michael Matloob <matloob@golang.org>
This field lists the names of non-Go files that are part of the
package under analysis.
Change-Id: Ic967dc18b98e018c691442f7378cb29db30a1454
Reviewed-on: https://go-review.googlesource.com/138136
Reviewed-by: Michael Matloob <matloob@golang.org>
Forcing clients to register their Fact types with gob made for an
unfriendly API. Now the driver is again responsible for doing it.
The FactTypes field is now a slice of Fact values (not reflect.Types)
used only for their dynamic type, which is slightly more convenient.
Change-Id: I01219edb24bd2371ba642bb56508aa80c19a9b61
Reviewed-on: https://go-review.googlesource.com/137836
Run-TryBot: Alan Donovan <adonovan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Matloob <matloob@golang.org>
...by removing references to trace API for now.
Change-Id: Ide6bbbfd98e15a3773b4a10232bcbf2dc2153341
Reviewed-on: https://go-review.googlesource.com/137615
Reviewed-by: Alan Donovan <adonovan@google.com>
Package checker is a driver for the analysis API.
It is an internal package, but is exposed by three different APIs:
analysistest, singlechecker, and multichecker.
Checker uses go/packages to load the specified packages (plus their
dependencies, if any analysis uses facts) from source code.
It constructs a graph of actions (analysis passes), whose dependency
edges may be "horizontal", when one analyzer depends on the output of
another applied to the same package, or "vertical", when an an
analyzer consumes facts produced by the same analyzer applied to a
dependency package.
The graph is executed in parallel, unless -debug=p.
Facts are passed from one pass to another in memory.
If -debug=s, facts are serialized, to exercise that logic.
Findings are printed at the end.
The -json flag selects JSON output.
Use -debug=t to print timing information.
Always use -debug=tp, for sequential mode, when timing.
Also:
- analysistest: a wrapper for testing checkers
- multichecker: a wrapper for writing multi-checker tools
Analysis flags are prefixed by name: -findcall.name=foo.
- cmd/analyze: a command-line tool based on multichecker
- singlechecker: a wrapper for writing single-checker tools
Analysis flags are unprefixed: -name=foo.
- passes/findcall/cmd/findcall: a standalone tool for the findcall analysis
- tests for findcall
- tests for pkgfact
Change-Id: Icfd4a49cee17e7de1ddb6ec15a62dc667fb2db04
Reviewed-on: https://go-review.googlesource.com/135679
Reviewed-by: Ian Cottrell <iancottrell@google.com>
Alan Donovan