This is the mail archive of the mailing list for the GDB project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: make_cleanup_func elimination

Jim Kingdon wrote:
> This patch takes the advice of the comments in defs.h.
> I did try a few testsuite runs without obvious disasters, but I needed
> to re-do them to get completely good numbers, and I haven't been
> getting around to that.  So sending the patch in now seemed better
> than just letting it sit around without telling anyone.
> 2000-03-13  Jim Kingdon  <>
>         Fix dubious assumptions about sizes of host pointers and ints and
>         such in calling make_cleanup:
>         * coffread.c, solib.c, dbxread.c, mipsread.c, os9kread.c,
>         elfread.c, nlmread.c, somread.c, xcoffread.c, ax-gdb.c,
>         tracepoint.c, dwarf2read.c, command.c, eval.c, top.c, event-top.c,
>         top.h, breakpoint.h, breakpoint.c, infrun.c, infcmd.c, varobj.c,
>         parse.c, stack.c, printcmd.c, symfile.c, symtab.c, thread.c,
>         typeprint.c, valops.c, procfs.c, hp-psymtab-read.c, hppa-tdep.c,
>         sparcl-tdep.c, remote-udi.c, procfs.c, remote-e7000.c: Don't cast
>         argument to make_cleanup.  Instead give the function we pass the
>         right argument type one way or another.  In various cases change
>         argument types or add wrapper functions.  In the case of bfd_close
>         (many places), close (procfs.c), closedir (procfs.c), chdir
>         (top.c), fclose (top.c), add error handling.
>         * defs.h (make_cleanup_func): Remove.


 (Yes, this will be hard to take seriously given that I botched the
original change).

The main thing to note is that the reason I didn't rush back and fix the
make_cleanup_func botch (ok so JasonM helped me :-) was because, as far
as anyone knows (gcc people here), while the code is technically in
violation of ISO-C there isn't a mainstream platform on which the hack
will actually fail. Consequently, there is no pressure to apply a quick
fix and instead changes that improve the overall code are being sort. 
This is why I've been slowly going through and adding functions like:

    extern struct cleanup *make_cleanup_ui_file_delete (struct ui_file

rather than changing functions signatures:

-static void free_variable PARAMS ((struct varobj * var));
+static void free_variable (void *);

The original function was strongly typed.  By changing it to void* that
type information is lost and the chance of incorrect use increased.

If you wan't to persue this then could I suggest breaking the patch down
into specific fixes and submitting each in turn, that way each
maintianer can merge it in.  Alternativly, I can file it into my post
5.0 branch list (1) and I'll work through them.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]