...since <https://github.com/llvm/llvm-project/commit/
d0da5d2bbe8305d06dc01a98706fd73e11e24a9f> "Change default traversal in AST
Matchers to ignore invisible nodes". This caused failures
> [CPT] compilerplugins/clang/test/constparams.cxx
> ParmVarDecl 0x11d76c730 <compilerplugins/clang/test/constparams.cxx:15:12, col:18> col:18 used f1 'int *'
> DeclRefExpr 0x11d76c948 'int *' lvalue ParmVar 0x11d76c730 'f1' 'int *'
> ParmVarDecl 0x11d76cc80 <compilerplugins/clang/test/constparams.cxx:21:12, col:18> col:18 used f2 'int *'
> DeclRefExpr 0x11d76ce60 'int *' lvalue ParmVar 0x11d76cc80 'f2' 'int *'
> error: 'error' diagnostics expected but not seen:
> File compilerplugins/clang/test/constparams.cxx Line 15: this parameter can be const Class1::Class1 [loplugin:constparams]
> error: 'error' diagnostics seen but not expected:
> File compilerplugins/clang/test/constparams.cxx Line 15: no parent? [loplugin:constparams]
> File compilerplugins/clang/test/constparams.cxx Line 21: no parent? [loplugin:constparams]
> 3 errors generated.
[...]
> [CPT] compilerplugins/clang/test/unusedenumconstants.cxx
> error: 'error' diagnostics expected but not seen:
> File compilerplugins/clang/test/unusedenumconstants.cxx Line 30: read Bottom [loplugin:unusedenumconstants]
> error: 'error' diagnostics seen but not expected:
> File compilerplugins/clang/test/unusedenumconstants.cxx Line 30: write Bottom [loplugin:unusedenumconstants]
> 2 errors generated.
[...]
> [CPT] compilerplugins/clang/test/unusedfields.cxx
> error: 'error' diagnostics expected but not seen:
> File compilerplugins/clang/test/unusedfields.cxx Line 156 (directive at compilerplugins/clang/test/unusedfields.cxx:164): write m_f5 [loplugin:unusedfields]
> File compilerplugins/clang/test/unusedfields.cxx Line 210 (directive at compilerplugins/clang/test/unusedfields.cxx:211): read m_f1 [loplugin:unusedfields]
> 2 errors generated.
For compilerplugins/clang/test/constparams.cxx at least it would have worked to
fix that locally with
> diff --git a/compilerplugins/clang/constparams.cxx b/compilerplugins/clang/constparams.cxx
> index 95c8184009d7..70f056fa5a69 100644
> --- a/compilerplugins/clang/constparams.cxx
> +++ b/compilerplugins/clang/constparams.cxx
> @@ -274,7 +274,7 @@ bool ConstParams::checkIfCanBeConst(const Stmt* stmt, const ParmVarDecl* parmVar
> {
> for ( auto cxxCtorInitializer : cxxConstructorDecl->inits())
> {
> - if ( cxxCtorInitializer->getInit() == stmt)
> + if ( cxxCtorInitializer->getInit()->IgnoreImpCasts() == stmt)
> {
> if (cxxCtorInitializer->isAnyMemberInitializer())
> {
(somewhat unintuitively, given the Clang change is apparently about ignoring
more implicit nodes), but overall it appears better---at least for now---to use
a getParents variant that keeps the old traversal behavior.
For that, instead of using the clang::ASTContext::ParentMapCtx, we create our
own loplugin::Plugin::parentMapContext_. There appear to be no uses of
ASTContext::getParent across the Clang codebase itself, outside of ASTMatcher
code, so it looks unlikely that creating our own ParentMapContext instance would
degrade performance by no longer sharing cached data between Clang's internals
and our plugin. (And given that ASTContext::getParents is deprecated with the
note "New callers should use ParentMapContext::getParents() directly", this may
well be the correct way in the long run, anyway.)
Change-Id: I46c7912f2737e7c224fd45ab41441f69e2f10bd4
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/94795
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
noticed something that wasn't being picked up, wrote some tests,
and found an unhandled case in Plugin::getParentFunctionDecl
Change-Id: I52b4ea273be6614e197392dfc4d6053bbc1704de
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/90141
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
...in <https://github.com/llvm/llvm-project/commit/
8a81daaa8b58aeaa192a47c4ce7f94b4d59ce082> "[AST] Split parent map traversal
logic into ParentMapContext.h", causing failures like
> compilerplugins/clang/constmethod.cxx: In member function ‘bool {anonymous}::ConstMethod::checkIfCanBeConst(const clang::Stmt*, const clang::CXXMethodDecl*)’:
> compilerplugins/clang/constmethod.cxx:191:70: error: invalid use of incomplete type ‘class clang::DynTypedNodeList’
> 191 | auto parentsRange = compiler.getASTContext().getParents(*stmt);
> | ^
> In file included from compilerplugins/clang/plugin.hxx:15,
> from compilerplugins/clang/constmethod.cxx:16:
> include/clang/AST/ASTContext.h:97:7: note: forward declaration of ‘class clang::DynTypedNodeList’
> 97 | class DynTypedNodeList;
> | ^~~~~~~~~~~~~~~~
Change-Id: Ib82d04608fa306a715af481422017c24053a01c6
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/87533
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
This is a continuation of ff002524c1.
When compiling with icecream, its -frewrite-includes merges
all #include's into one .cxx file and marks them with with line
markers. But SourceManager::getFilename() doesn't take those
into account and so it reports all of those as <stdin>.
So use getFileNameOfSpellingLoc(), which explicitly handles this
case.
And we should probably never ever use SourceManager::getFilename().
Change-Id: Ia194c2e041578e1e199aee2df2f885922ef7e31a
Reviewed-on: https://gerrit.libreoffice.org/80326
Tested-by: Jenkins
Reviewed-by: Luboš Luňák <l.lunak@collabora.com>
treat fields touched in operator== as not being important, which
finds some more stuff (but also adds some false+)
Change-Id: I3f5d504d7dec7945a917afbcd58c92df74f03645
Reviewed-on: https://gerrit.libreoffice.org/62020
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
look for collection-like fields that are never added to, and are
therefore effectively unused
Change-Id: Id52c5500ea5e3d2436fb5915aebb86278bf2d925
Reviewed-on: https://gerrit.libreoffice.org/60661
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
look for fields which are only assigned to in the constructor, so they
can be made const
Change-Id: I0b76817c2181227b04f6a29d6a808f5e31999765
Reviewed-on: https://gerrit.libreoffice.org/60393
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
by checking if the current namespace decl is in our code, so we have to
scan less stuff, which results in a 10% perf improvement for me
Change-Id: Idf0e30d57b6d0dcd13daa9ed679c28b9d233d387
Reviewed-on: https://gerrit.libreoffice.org/58942
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
...which first added alternative names to and then deprecated getLocBegin/End
Change-Id: Iaefb8ce259057abfa6cd20f0b63c0ef2949a96b2
Reviewed-on: https://gerrit.libreoffice.org/58820
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
idea from tml.
Extend the unusedfields plugin to find fields that are only assigned in
the constructor.
Change-Id: I258d3581afbe651d53ce730c9ba27a4598cd9248
Reviewed-on: https://gerrit.libreoffice.org/57733
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
so that the next time I accidentally leave one turned on, and commit it,
the buildbots will clean up naturally the next time they run 'make
clean'
Change-Id: Ia09dea9c272c322c7e2773c5458cb54aceb50dd1
improve the read-only check to ignore reads from fields that are guarded
by a boolean check, something like:
if (field)
field.foo();
this produces some false positives at the moment because I'm not
correctly handling the else block, but also some useful new dead code.
Change-Id: Id21fa1a56c171d09d979769b978b6eef14e8b695
Reviewed-on: https://gerrit.libreoffice.org/52664
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
for some reason we're hitting more template AST nodes now? Anyhow,
updated singlevalfields and unusedenumconstants to cope.
For unusedfields, ignore field access inside Clone() methods, since it's
like a constructor.
Similarly for unusedmethods.
Change-Id: Icb2f76fb2f06ae5df21f9d75312e42a2800befb9
Reviewed-on: https://gerrit.libreoffice.org/45470
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
to getParentStmt and rename parentFunctionDecl to getParentFunctionDecl,
so I don't keep using accidentally naming my variables the same as the
functions.
Change-Id: I66f9452458c8b439e5132191ac5219fb6d420708
which makes absolutely no difference to the results, but anyhow, would
be a shame to waste the work
Change-Id: I4576528f30986a5ce522c76fdf21873f0ce23f0a
(*) IsPassedByNonConst was completely wrong, not even sure why it worked
before.
(*) treat a field passed to operator>>= as being written to, but not
read
Change-Id: Id3a5f2f35222986fe5edba3f5a58215a1815d401
(*) better analysis of init-list-expressions
(*) fix analysis of calls to members, turns out there is no parameter
offset after all
(*) check for passing arrays to functions, need to check
if the parameter is T* or T const *
(*) check for assigning field to a T& variable
Change-Id: Ie6f07f970310c3854e74619fe4fd02a299bf6879
found a couple more by running
make build-nocheck
to exclude testing code
Also, teach unusedfields loplugin about operator<<, since
referring to a field in an ostream operator<< does not indicate a
real use, it's normally for debugging
Change-Id: I3dce22bf5afda2fd09e01de9bf1d0ef85e535aa8
Reviewed-on: https://gerrit.libreoffice.org/39625
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
for some reason the insideMoveOrCopyDecl pointer to MethodDecl becomes
bad during AST traversal, but the pointers to RecordDecl seem stable?
Change-Id: Ida939f5ca4780e674b245411f7395f147258544e
