large symbol file support

[derekbruening]

From derek.br...@gmail.com on December 10, 2010 17:58:24

PR 583343

for symbol files that don't fit in memory (some combination of large or
many symbol files and significant memory usage by the application) the plan
is to use sideline processing which needs support from issue #44/PR 243532.

may also want to support the post-processing model that is now supported on
Linux via two-step -skip_results/-results invocations (PR 575481)

xref PR 540913: online symbol processing
xref issue #44/PR 243532: [API] symbol table lookup support

Original issue: http://code.google.com/p/drmemory/issues/detail?id=192

[derekbruening]

From bruen...@google.com on September 30, 2011 07:28:00

on Linux as well as Cygwin the plan is to switch from sideline to online symbol processing and unify the two platforms, once we have DRi#445.

we still want post-processing support on all platforms ( issue #446 ) for re-running w/ better symbols or w/ new suppressions. and w/ symbol cache support, a different model for supporting too-large-to-map symbol files could be to generate a symbol cache file up front for correctness of the tool and then use post-processing for callstacks. this would mean we would not need any sideline support at all.

Summary: large symbol file support
Labels: -OpSys-Windows

[derekbruening]

From bruen...@google.com on September 30, 2011 07:34:00

symbol cache is issue #604

[derekbruening]

From bruen...@google.com on December 20, 2011 15:14:01

adding a note that anyone synthetically created a symcache needs to be aware of wildcard symcache entries: issue #722 added "std::_DebugHeapDelete<*>" whose matches are stored as
"std::_DebugHeapDelete<>" duplicates.

[derekbruening]

From bruen...@google.com on March 13, 2012 11:25:36

xref list of reasons we need symbols in issue #448 no support for -results_to_stderr? but, only have to avoid one module.
rest can use online syms. so could still have -results_to_stderr, just
missing syms for executable or main dll, and can go look afterward and line
it up (via error#)

to get real syms would have to pass all modules w/ static libc up front:
but again, can have online syms for all but the giant module