RE: [PING] [PATCH] gdb: cooked_index_functions::expand_symtabs_matching: require psymtab

RE: [PING] [PATCH] gdb: cooked_index_functions::expand_symtabs_matching: require psymtab

[Six, Lancelot]

[AMD Official Use Only - General]

Kindly pinging

Best,
Lancelot.

-----Original Message-----
From: Six, Lancelot 
Sent: 04 May 2022 18:12
To: Tom Tromey <tom@tromey.com>; Lancelot SIX via Gdb-patches <gdb-patches@sourceware.org>
Cc: lsix@lancelotsix.com
Subject: RE: [PATCH] gdb: cooked_index_functions::expand_symtabs_matching: require psymtab

[AMD Official Use Only - General]

Hi Tom,

I did create a branch where I completely removed the lazy loading of partial symtabs (I can push it somewhere if this is of interest).

I have experimented with a scenario where I run `make` under GDB in a project using a debug build of Clang.  Using the lazy loading, everything runs in ~15 minutes (without parallel building), while if I deactivate the lazy loading the build takes ~50 minutes.  So the lazy loading definitely brings improvement for some scenarios.

Another test I did was to run the testsuite with the patch I propose, plus another one forcing lazy-lading as much as possible.  I do have some regressions, but most of them are linked to information messages appearing when the test does not expect them (or failing to appear when the test expect them like after using the 'file' command).  I did not investigate all the regressions, but the important point is that if I do not have the change in cooked_index_functions::expand_symtabs_matching I see GDB crash in many tests, while with this change it does not.

Best,
Lancelot.

-----Original Message-----
From: Tom Tromey <tom@tromey.com>
Sent: 26 April 2022 20:29
To: Lancelot SIX via Gdb-patches <gdb-patches@sourceware.org>
Cc: Six, Lancelot <Lancelot.Six@amd.com>; lsix@lancelotsix.com
Subject: Re: [PATCH] gdb: cooked_index_functions::expand_symtabs_matching: require psymtab

[CAUTION: External Email]

>>>>> Lancelot SIX via Gdb-patches <gdb-patches@sourceware.org> writes:

> The new indexer's expand_symtabs_matching callback does not have a 
> call to the objfile's require_partial_symbols, so there is not 
> mechanism to ensure that the psymtab is laded at this point.

I wonder if this lazy loading is even worthwhile any more.
Perhaps just removing SYMFILE_NO_READ entirely is the way to go.

Tom