This change adds rwlocks.rlock and rwlocks.wlock symlinks rwlocks bin… #205
Conversation
|
@Sashan please reconsider this approach. IMO it is not scaleable. IMO we need to be able to pass some arbitrary options from the statistics collection script and be able to invoke the same perf tool with different options. |
|
I agree it's not scalable. On the other hand it is good enough for what we have now. We just glue it with script I got from Niel collect the results almost automatically and be done with it for 3.4. We can then revisit it later. In this particular case the It's true we can make the script which drives performance tests more complex so it will be able to deal with multiline output but this really clobbers the logic in loop which executes the individual tools. Right now I'm looking for some simple and quick solution which will get us going. We can always revisit later, in my opinion. |
|
I have no problem with adding an option for limiting the output to just one number. What I have a problem with is the creation of the duplicate binaries (or symlinking) instead of having a possibility to specify extra arguments in the statistics collection script. |
…ary. We need it so we can pass 'implicit' option so we can chose which statistics want to collect. This helps us with our current simple tooling we have to collect stats. The script we currently work on expects tool prints a single number as its output.
|
Agreed @t8m, options are better here. |
I disagree. Please refer to script The extra/special options would ask us to add logic into the loop which runs all tools for all library versions. In my opinion we are winning if we don't need those special options. I see this way as good enough for now so we can use the same script to collect data we need for unix-like systems. And I agree the approach has limits in longrun, but I think we can deal with those later as our test performance infrastructure will grow. Right now I'd like to start with something minimalistic just to start moving. |
|
I think adding an explicit list of commands with arguments into the perf.sh script instead just list of commands is nothing that would complicate the run-perf.sh script in any serious way. |
the reason why I persist on the idea of implicit options (at least for now) is that we can use wildcards to selectively chose tests to run. for example if we creat yeat-another custom script to get numbers quickly for rwlocks and pkeyread in ber format we just do: if we will be using explicit options to I'm also not saying the implicit options are perfect or ultimate thing we should keep forever. I see it as good enough for now just to get going. At this point we have a simple script which seems to be doing its job. We can always go for more fancy solution with explicit options later. but if you really insist I can give it a try and craft |
|
I also prefer doing this using arguments. What's here is Unix specific which precludes running these performance tests on non-Unixy platforms. I appreciate the wildcard concern but don't find it compelling. |
of implicit arguments (symlinks to tools). The approach is discussed in those two PRs: openssl#205 openssl#203
|
Please move this PR to perftools repo if it is still relevant. |
…ary.
We need it so we can pass 'implicit' option so we can chose which statistics want to collect. This helps us with our current simple tooling we have to collect stats. The script we currently work on expects tool prints a single number as its output.