27 lines
969 B
Plaintext
27 lines
969 B
Plaintext
|
Things that still need to be done: -*- Text -*-
|
||
|
|
|
||
|
|
o - Performance
|
||
|
|
|
||
|
|
All performance could be tuned, but one area that could be looked
|
||
|
|
at especially is performance with flags, particularly
|
||
|
|
--detect-odr-violations and --compress-debug-sections.
|
||
|
|
|
||
|
|
o - Threads
|
||
|
|
|
||
|
|
Why is the usertime when we run with threads the same (or almost
|
||
|
|
the same) as when we run without? Is it because threads spend most
|
||
|
|
of their time waiting on the same resources? On each other?
|
||
|
|
Something else?
|
||
|
|
|
||
|
|
o - ODR false positives
|
||
|
|
|
||
|
|
ODR false positives can happen when we optimize, since code in .h
|
||
|
|
files may be optimized in different ways in different compilation
|
||
|
|
units. It's possible we could fix this for real by looking at the
|
||
|
|
full debug info and using DW_TAG_inlined_subroutine in a clever way
|
||
|
|
to correct for inlining. But that would be very expensive, I
|
||
|
|
think. The easier solution is to recommend people only do
|
||
|
|
ODR-detection with -g0.
|
||
|
|
|
||
|
|
o - Better testing
|