Re: [PATCH v2] Capture warnings when writing to the index cache

[Simon Marchi <simark@simark.ca>]

On 2024-03-26 11:49, Tom Tromey wrote:
>>>>>> "Tom" == Tom Tromey <tromey@adacore.com> writes:
> 
> Tom> PR symtab/30837 points out a race that can occur when writing to the
> Tom> index cache: a call to ada_encode can cause a warning, which is
> Tom> forbidden on a worker thread.
> 
> Tom> This patch fixes the problem by arranging to capture any such
> Tom> warnings.
> 
> Tom> This is v2 of the patch.  It is rebased on top of some other changes
> Tom> in the same area.  v1 was here:
> 
> Tom>     https://sourceware.org/pipermail/gdb-patches/2024-February/206595.html
> 
> Tom> Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30837
> 
> I'm checking this in now.
> 
> Tom

Hi Tom,

When building with Clang 17, I see:

    /home/simark/src/binutils-gdb/gdb/../gdbsupport/function-view.h:300:37: error: format string is not a string literal [-Werror,-Wformat-nonliteral]
      300 |       noexcept (noexcept (callable (std::forward<Args> (args)...))) -> Res
          |                                     ^~~~~~~~~~~~~~~~~~~~~~~~~
    /home/simark/src/binutils-gdb/gdb/../gdbsupport/function-view.h:299:17: note: while substituting into a lambda expression here
      299 |     m_invoker = [] (fv_detail::erased_callable ecall, Args... args)
          |                 ^
    /home/simark/src/binutils-gdb/gdb/../gdbsupport/function-view.h:265:5: note: in instantiation of function template specialization 'gdb::function_view<void (const char *, __va_list_tag *)>::bind<deferred_warnings>' requested here
      265 |     bind (callable);
          |     ^
    /home/simark/src/binutils-gdb/gdb/dwarf2/cooked-index.c:592:42: note: in instantiation of function template specialization 'gdb::function_view<void (const char *, __va_list_tag *)>::function_view<deferred_warnings &, void, void>' requested here
      592 |       scoped_restore_warning_hook defer (*warn);
          |                                          ^
    In file included from <built-in>:1:
    In file included from /home/simark/src/binutils-gdb/gdb/defs.h:620:
    In file included from /home/simark/src/binutils-gdb/gdb/utils.h:24:
    /home/simark/src/binutils-gdb/gdb/../gdbsupport/function-view.h:305:34: error: format string is not a string literal [-Werror,-Wformat-nonliteral]
      305 |         return (Res) restored_callable (std::forward<Args> (args)...);
          |                                         ^~~~~~~~~~~~~~~~~~~~~~~~~

My interpretation of this is that we bind a function marked with
`ATTRIBUTE_PRINTF (2,0)` (deferred_warnings::operator()) to a function
view that doesn't convey the attribute.  So I guess the compiler is
telling us that whenever the function view is invoked, there won't be a
format string check, and that's dangerous.

I didn't find a good fix that :(.

Simon