[Bug symtab/29105] new DWARF reader still slow

["simon.marchi at polymtl dot ca" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=29105

--- Comment #7 from Simon Marchi <simon.marchi at polymtl dot ca> ---
I'm currently debugging objfile::lookup_symbol, and it seems the intention to
stop expanding when the symbol is found.  What I see is that we are looking for
the "signed int" type:

(top-gdb) frame
#6  0x0000555555c5dd69 in objfile::lookup_symbol (this=0x555556807040,
kind=STATIC_BLOCK, name=0x7fffffffaca0 "signed int", domain=VAR_DOMAIN) at
symfile-debug.c:276
276           if (!iter->expand_symtabs_matching (this,

But in cooked_index_functions::expand_symtabs_matching, we decided to match
index entries with "int" instead of "signed int" (see the name_vec vector):

(top-gdb) frame
#5  0x00005555558e1d07 in
cooked_index_functions::expand_symtabs_matching(objfile*,
gdb::function_view<bool (char const*, bool)>, lookup_name_info const*,
gdb::function_view<bool (char const*)>, gdb::function_view<bool
(compunit_symtab*)>, enum_flags<block_search_flag_values>, domain_enum_tag,
search_domain) (this=0x555556809360, objfile=0x555556807040, file_matcher=...,
lookup_name=0x7fffffffa760, symbol_matcher=..., expansion_notify=...,
search_flags=..., domain=VAR_DOMAIN, kind=ALL_DOMAIN) at dwarf2/read.c:20288
20288             if (!dw2_expand_symtabs_matching_one (entry->per_cu,
per_objfile,
(top-gdb) p name_vec
$13 = std::vector of length 1, capacity 1 = {"int"}

Most CUs have a type called "int", so that matches a lot of them.

Back in the expansion notifier in objfile::lookup_symbol, we search the just
expanded CUs for the "signed int" type, and we don't find it, so we return
false, meaning "not found, please expand the next CU".

I'm not too familiar with this code, but the discrepancy between the criteria
on which we choose the CUs to expand (those that have the "int" type) and then
the type we actually look for ("signed int") is odd.

In the previous psymtabs code, I think we looked for partial symbols named
"signed int", and didn't find any, so it didn't cause expansion of these CUs.

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