efriedma added inline comments.
================
Comment at: clang/test/Sema/warn-vla.c:8-12
+void test2(int n, int v[n]) { // c99 no-warning
+#if __STDC_VERSION__ < 199901L
+// expected-warning@-2{{variable length arrays are a C99 feature}}
+#endif
}
----------------
aaron.ballman wrote:
> efriedma wrote:
> > aaron.ballman wrote:
> > > aaron.ballman wrote:
> > > > inclyc wrote:
> > > > > aaron.ballman wrote:
> > > > > > The diagnostic there is rather unfortunate because we're not using
> > > > > > a variable-length array in this case.
> > > > > Emm, I'm not clear about whether we should consider this a VLA, and
> > > > > generates `-Wvla-extensions`. Is `v[n]` literally a variable-length
> > > > > array? (in source code) So it seems to me that we should still report
> > > > > c89 incompatibility warnings?
> > > > >
> > > > C89's grammar only allowed for an integer constant expression to
> > > > (optionally) appear as the array extent in an array declarator, so
> > > > there is a compatibility warning needed for that. But I don't think we
> > > > should issue a warning about this being a VLA in C99 or later. The
> > > > array *is* a VLA in terms of the form written in the source, but C
> > > > adjusts the parameter to be a pointer parameter, so as far as the
> > > > function's type is concerned, it's not a VLA (it's just a
> > > > self-documenting interface).
> > > >
> > > > Because self-documenting code is nice and because people are worried
> > > > about accidental use of VLAs that cause stack allocations (which this
> > > > does not), I think we don't want to scare people off from this
> > > > construct. But I'm curious what others think as well.
> > > > But I'm curious what others think as well.
> > >
> > > (Specifically, I'm wondering if others agree that the only warning that
> > > should be issued there is a C89 compatibility warning under
> > > `-Wvla-extensions` when in C89 mode and via a new `CPre99Compat`
> > > diagnostic group when enabled in C99 and later modes.)
> > >
> > >
> > I imagine people working with codebases that are also built with compilers
> > that don't support VLAs would still want some form of warning on any VLA
> > type.
> The tricky part is: that's precisely what WG14 N2992
> (https://www.open-std.org/jtc1/sc22/wg14/www/docs/n2992.pdf) is clarifying.
> If your implementation doesn't support VLAs, it's still required to support
> this syntactic form. So the question becomes: do we want a portability
> warning to compilers that don't conform to the standard? Maybe we do (if we
> can find evidence of compilers in such a state), but probably under a
> specific diagnostic flag rather than -Wvla.
That only applies to C23, though, right? None of that wording is there for
C11. (In particular, MSVC claims C11 conformance without any support for VLA
types.)
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D132952/new/
https://reviews.llvm.org/D132952
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
