[RFA take 5] Allow setting breakpoints on inline functions (PR 10738)

[Jan Kratochvil]

On Mon, 20 Feb 2012 16:58:49 +0100, Gary Benson wrote:
> This version of the patch has been updated to add a new command line
> option, --allow-incomplete-gdb-indexes,

Sorry I was a bit offline, I believe such option would be OK just as a vendor
patch.  Still I understand patching back the backward support is not easy so
OK, one day it can get finally dropped.


> --- a/gdb/dwarf2read.c
> +++ b/gdb/dwarf2read.c
[...]
> @@ -2121,13 +2125,20 @@ dwarf2_read_index (struct objfile *objfile)
>    /* Versions earlier than 3 emitted every copy of a psymbol.  This
>       causes the index to behave very poorly for certain requests.  Version 3
>       contained incomplete addrmap.  So, it seems better to just ignore such
> -     indices.  Index version 4 uses a different hash function than index
> -     version 5 and later.  */
> -  if (version < 4)
> +     indices.
> +
> +     Index version 4 uses a different hash function than index version
> +     5 and later.
> +
> +     Versions earlier than 5 did not emit psymbols for inlined

than 6

> +     functions.  Using these files will cause GDB not to be able to
> +     set breakpoints on inlined functions by name, so we ignore these
> +     indices unless --allow-incomplete-gdb-indexes was supplied.  */
> +  if (version < (allow_incomplete_gdb_indexes ? 4 : 6))
>      return 0;
>    /* Indexes with higher version than the one supported by GDB may be no
>       longer backward compatible.  */
> -  if (version > 5)
> +  if (version > 6)
>      return 0;
>  
>    map = OBSTACK_ZALLOC (&objfile->objfile_obstack, struct mapped_index);
[...]
> --- a/gdb/main.c
> +++ b/gdb/main.c
> @@ -87,6 +87,11 @@ int return_child_result_value = -1;
>  /* Whether to enable writing into executable and core files.  */
>  extern int write_files;
>  
> +/* Whether to use incomplete .gdb-index sections.
> +   The default is to reject them.  */
> +
> +extern int allow_incomplete_gdb_indexes;
> +

Please no "extern var" in *.c files, it may lead to non-matching types etc.
I understand it is present in the current codebase but this is not an excuse,
fixed it now:
	[obv] Code cleanup: main.c declarations
	http://sourceware.org/ml/gdb-patches/2012-02/msg00410.html

Put it to symfile.h where are declarations for other dwarf2read.c definitions.


> @@ -975,6 +982,10 @@ This is the GNU debugger.  Usage:\n\n\
>  Options:\n\n\
>  "), stream);
>    fputs_unfiltered (_("\
> +  --allow-incomplete-gdb-indexes\n\
> +                     Do not reject incomplete .gdb-index sections.\n\

I do not understand the option meaning in such terse form.  As English
non-native I would prefer just "old" everywhere (or Doug says "deprecated").

                        Do not reject old (incomplete) .gdb-index sections.\n\

OK for check-in with these and Eli's changes.


Thanks,
Jan