- We either need to move to a newer zope, or apply a small patch I have to our zope.app.publication (a backport of the two above patches) : I think its a matter of 'make a new egg in the sourcedeps, edit the versions list to use it', but I'm not sure of the QA process used on changes at that level.
- I have a patch to profiling which then uses this event and makes it easier to reason about.
- my uniquefileallocator patch passes all tests and will land once PQM opens again; that provides the basis for writing the file to disk Just-In-Time
- bzrlib's lsprof wrapper is ready to provide mutex support for this - but we can just ignore that in the short term, things will be tolerable ;)
- I, or someone else, needs to move profiling from per-thread to per-request using uniquefileallocator
- And finally convert to using callgrind files.
All in all only a couple more hours work, but longer than that due to various latencies. I'll keep this bug apprised of my progress.
Things that can help move it along: /bugs.edge. launchpad. net/zope. app.publication /+bug/598816 and /code.edge. launchpad. net/~lifeless/ zope.publisher/ start-request- event/+ merge/28558 - they seem stalled even though Tres was happy with them.
- help me land https:/
https:/
- We either need to move to a newer zope, or apply a small patch I have to our zope.app. publication (a backport of the two above patches) : I think its a matter of 'make a new egg in the sourcedeps, edit the versions list to use it', but I'm not sure of the QA process used on changes at that level.
- I have a patch to profiling which then uses this event and makes it easier to reason about.
- my uniquefileallocator patch passes all tests and will land once PQM opens again; that provides the basis for writing the file to disk Just-In-Time
- bzrlib's lsprof wrapper is ready to provide mutex support for this - but we can just ignore that in the short term, things will be tolerable ;)
- I, or someone else, needs to move profiling from per-thread to per-request using uniquefileallocator
- And finally convert to using callgrind files.
All in all only a couple more hours work, but longer than that due to various latencies. I'll keep this bug apprised of my progress.