Hello,

In dwarf2read.c we sort global symbols automatically during the read. This, on 
large projects, causes long delay at startup.

For example, a project our customer has is a C++ project and is fairly large. 
Here is some statistics to give you an idea:

Statistics:
     Number of psymtabs         :    16659
     Number of global symbols   :  9134742
     Number of static symbols   :  1021345

The attached patch sorts on first lookup instead of on load.

Startup time (i.e. time until I get control when stopped in main) was before the 
patch:
   %   cumulative   self              self     total
  time   seconds   seconds    calls   s/call   s/call  name
  65.93     21.38    21.38       16     1.34     1.34  strcmp_iw_ordered
   5.70     23.23     1.85 10765999     0.00     0.00  bcache_data
...
And after the patch:
   %   cumulative   self              self     total
  time   seconds   seconds    calls   s/call   s/call  name
  20.43      1.80     1.80 10765999     0.00     0.00  bcache_data
  11.80      2.84     1.04 11137416     0.00     0.00  htab_hash_string
   8.74      3.61     0.77 14851934     0.00     0.00  htab_find_slot_with_hash
...

Note that 'strcmp_iw_ordered' comes from sort_pst_symbols (and indirectly from 
qsort); the number '16' in 'calls' column is due to wrap-around. Number of calls 
is much higher.

Testing on linux, no regressions.

---
Aleksandar Ristovski
QNX Software Systems


ChangeLog
2008-02-11  Aleksandar Ristovski  <aristovski@qnx.com>

	* dwarf2read.c (dwarf2_build_psymtabs_hard): Remove call to
	sort_pst_symbols. It will be done on first lookup.
	* symfile.c (compare_psymbols): Set globals_sorted flag.
	* symmisc.c (maintenance_info_psymtabs): Add some statistics to the
	output.
	* symtab.c (lookup_partial_symbol): Sort global symbols if not
	already sorted, then go on with the appropriate lookup.
	* symtab.h (struct partial_symtab): Add new field globals_sorted.