[Bug gdb/31636] New: GDB 14 incorrectly reads/handles file timestamps on i686-w64-mingw32 hosts

[Bug gdb/31636] New: GDB 14 incorrectly reads/handles file timestamps on i686-w64-mingw32 hosts

[peter0x44 at disroot dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31636

            Bug ID: 31636
           Summary: GDB 14 incorrectly reads/handles file timestamps on
                    i686-w64-mingw32 hosts
           Product: gdb
           Version: 14.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: gdb
          Assignee: unassigned at sourceware dot org
          Reporter: peter0x44 at disroot dot org
  Target Milestone: ---

When using gdb on an i686-w64-mingw32 host, GDB always detects the executable
as outdated. As a consequence, it fails to detect file changes and is unusable.

This is a regression between gdb 13 and 14, and gdb 14.2 did not fix this.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug gdb/31636] GDB 14 incorrectly reads/handles file timestamps on i686-w64-mingw32 hosts

[peter0x44 at disroot dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31636

Peter Damianov <peter0x44 at disroot dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |i686-w64-mingw32
                URL|                            |https://github.com/skeeto/w
                   |                            |64devkit/issues/96

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug gdb/31636] GDB 14 incorrectly reads/handles file timestamps on i686-w64-mingw32 hosts

[peter0x44 at disroot dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31636

--- Comment #1 from Peter Damianov <peter0x44 at disroot dot org> ---
The build of GDB was cross-compiled using Debian bookworm under docker.
https://github.com/skeeto/w64devkit/blob/deef5631c9ebeb5f401e5744ee3b4d22ed56a3f6/Dockerfile#L357-L368

I believe --host=i686-w64-mingw32 should be enough to reproduce, but I haven't
tried.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug gdb/31636] GDB 14 incorrectly reads/handles file timestamps on i686-w64-mingw32 hosts

[ssbssa at sourceware dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31636

Hannes Domani <ssbssa at sourceware dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ssbssa at sourceware dot org

--- Comment #2 from Hannes Domani <ssbssa at sourceware dot org> ---
The reason is, gnulib/config.h has this define:

```
/* For 64-bit time_t on 32-bit mingw. */
#define __MINGW_USE_VC2005_COMPAT 1
```

And bfd doesn't do this, so you get different definitions of struct stat.

In bfd/bfdio.c:

```
(top-gdb) pt buf
type = struct _stati64 {
    _dev_t st_dev;
    _ino_t st_ino;
    unsigned short st_mode;
    short st_nlink;
    short st_uid;
    short st_gid;
    _dev_t st_rdev;
    long long st_size;
    __time32_t st_atime;
    __time32_t st_mtime;
    __time32_t st_ctime;
}
```

And in gdb/symfile.c:

```
(top-gdb) pt new_statbuf
type = struct _stat64 {
    _dev_t st_dev;
    _ino_t st_ino;
    unsigned short st_mode;
    short st_nlink;
    short st_uid;
    short st_gid;
    _dev_t st_rdev;
    long long st_size;
    __time64_t st_atime;
    __time64_t st_mtime;
    __time64_t st_ctime;
}
```

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug gdb/31636] GDB 14 incorrectly reads/handles file timestamps on i686-w64-mingw32 hosts

[pedro at palves dot net]

https://sourceware.org/bugzilla/show_bug.cgi?id=31636

Pedro Alves <pedro at palves dot net> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
                 CC|                            |pedro at palves dot net
   Last reconfirmed|                            |2024-04-12
     Ever confirmed|0                           |1

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug gdb/31636] GDB 14 incorrectly reads/handles file timestamps on i686-w64-mingw32 hosts

[peter0x44 at disroot dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31636

--- Comment #3 from Peter Damianov <peter0x44 at disroot dot org> ---
Can confirm that adding -D__MINGW_USE_VC_2005_COMPAT in CFLAGS and CXXFLAGS
appears to resolve this issue.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug gdb/31636] GDB 14 incorrectly reads/handles file timestamps on i686-w64-mingw32 hosts

[peter0x44 at disroot dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31636

--- Comment #4 from Peter Damianov <peter0x44 at disroot dot org> ---
I meant to write:
-D__MINGW_USE_VC2005_COMPAT

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug gdb/31636] GDB 14 incorrectly reads/handles file timestamps on i686-w64-mingw32 hosts

[ssbssa at sourceware dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31636

--- Comment #5 from Hannes Domani <ssbssa at sourceware dot org> ---
(In reply to Peter Damianov from comment #3)
> Can confirm that adding -D__MINGW_USE_VC_2005_COMPAT in CFLAGS and CXXFLAGS
> appears to resolve this issue.

Yes, I would have suggested a similar workaround for now (though using CPPFLAGS
instead of CFLAGS/CXXFLAGS).

I was wondering why this problem didn't appear in gdb-13 as well, it already
contained this define since this gnulib update:
https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=dc6c21dabfbd242616b9e6266b459eded2d0762b

It's because reread_symbols() of gdb/symfile.c changed from stat() to
bfd_stat() here:
https://sourceware.org/git/?p=binutils-gdb.git;a=commitdiff;h=f2c4f78c813a9cef38b7e9c9ad18822fb9e19345

Should bfd be changed to always define __MINGW_USE_VC2005_COMPAT for mingw
targets like gnulib does?

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug gdb/31636] GDB 14 incorrectly reads/handles file timestamps on i686-w64-mingw32 hosts

[tromey at sourceware dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=31636

Tom Tromey <tromey at sourceware dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |tromey at sourceware dot org

--- Comment #6 from Tom Tromey <tromey at sourceware dot org> ---
I think we didn't anticipate that 'struct stat' itself could
vary between subdirectories like this.

Calling bfd_stat is unavoidable and that symfile.c change was
needed to fix a different bug -- at the very least that code
would fail for remote BFDs.

I wonder if there's a hack we could add to gnulib to avoid
some of its 'stat'-related changes.  Maybe that would be
sufficient.

Otherwise I think the only two solutions here are (1) also
get BFD to use gnulib, and (2) remove gnulib.

(1) is a bit of a pain because gdb has to carefully curate
exactly which gnulib modules are used -- since IIRC there's
at least one that we accidentally used and it broke other
things.

Probably (2) is best for our long-term sanity.  And maybe simpler
now that gdb uses c++ and can perhaps defer a subset of
the portability problems to the standard library.

-- 
You are receiving this mail because:
You are on the CC list for the bug.