objdump can take an extremely long time

[mikebentley15]

When doing symbol bisect, the symbol space to search over is generated using objdump --disassemble-all --line-numbers. For large object files with many symbols, especially if it is unoptimized, can take a very long time (such as minutes for a single invocation and take up 50 MB of memory or more). There should be a better way to extract the line number of functions rather than doing a full disassembly.

This is an optimization that deals with a worst case scenario. It is unknown how often this worst case scenario is encountered, and therefore, the impact on the potential community of users is unknown. If you have experienced this problem with FLiT bisect, please chime in.

[mikebentley15]

What about using CTags to get function line numbers instead? Yes, this increases to one more dependency, but not a big one, not a required one, and it should be much more performant.

See python-ctags for accessing the ctags information from python.

[mikebentley15]

cscope is another possability

[mikebentley15]

Fixed by pull request #241