Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

please add an option to enable debug symbols only #6238

Open
vicuna opened this Issue Nov 14, 2013 · 7 comments

Comments

Projects
None yet
2 participants
@vicuna
Copy link
Collaborator

vicuna commented Nov 14, 2013

Original bug ID: 6238
Reporter: @ygrek
Assigned to: @mshinwell
Status: acknowledged (set by @mshinwell on 2013-11-18T16:29:38Z)
Resolution: open
Priority: normal
Severity: feature
Target version: undecided
Category: back end (clambda to assembly)
Related to: #6728
Monitored by: @gasche @ygrek @yakobowski @alainfrisch

Bug description

Rationale:
currently option -g mixes several things up - generation debug symbols, storing backtrace information, adding more precise bounds checking, disabling some optimizations (in bytecode).
For profiling one usually wants the same binary code that is run in production but with debug symbols. So the speed of the code is the same, but there is some extra information available on demand, and this build can be run on production at highest speed and then profiled live if any problem arises. Option -g makes this approach less attractive because of incurred performance cost (e.g. storing backtrace).
Can we please have another option to only write out (native, i.e. DWARF) debugging info and perform no other changes to generated code?

@vicuna

This comment has been minimized.

Copy link
Collaborator Author

vicuna commented Nov 18, 2013

Comment author: @mshinwell

I think this is a reasonable thing to do. Watch this space.

@vicuna

This comment has been minimized.

Copy link
Collaborator Author

vicuna commented Nov 18, 2013

Comment author: @gasche

While I think this is a reasonable feature, I think that in the long term the way to go is rather to provide more control to performance-conscious users of which raises will trigger backtrace recording (eg. the raise-variants work of Alain). Indeed, there should be a sweet spot where most exceptions used for control flow do not record any trace, only actually exceptional exceptions do, and you have a system that is both efficient (essentially as the no-trace one) and auditable in case of unplanned failure (a good reason to have some backtraces even in production). I can understand people disabling everything for absolute performances (and right now it's probably easier to implement), but long-term I think that's more of a minority use-case.

Of course if someone is ready to do the work, that's great in any case.

@vicuna

This comment has been minimized.

Copy link
Collaborator Author

vicuna commented Feb 19, 2014

Comment author: @damiendoligez

Gabriel, your argument addresses backtraces only, not the bound-checking and disabled optimizations.

@vicuna

This comment has been minimized.

Copy link
Collaborator Author

vicuna commented Dec 23, 2014

Comment author: @alainfrisch

more precise bounds checking

Have you seen cases where keeping multiple call sites for caml_ml_array_bound_error yields a noticeable performance penalty?

disabling some optimizations (in bytecode)

Since the request is to have a mode with maximal runtime performance (while keeping debugging information), I don't think that bytecode is relevant.

Alain

@vicuna

This comment has been minimized.

Copy link
Collaborator Author

vicuna commented Jan 9, 2015

Comment author: @damiendoligez

In any case, the request is for adding debugging info to the executable, while guaranteeing no run-time overhead, and this seems quite reasonable to me.

@vicuna

This comment has been minimized.

Copy link
Collaborator Author

vicuna commented Apr 18, 2016

Comment author: @gasche

Mark, is it time for a gentle ping?

@vicuna

This comment has been minimized.

Copy link
Collaborator Author

vicuna commented Jun 9, 2016

Comment author: @mshinwell

Please see the comment on the pull request

#574

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.