4

I would like xdebug to trace only "jumps" of over X ms or of over Y KB RAM, for instance, every execution that took longer than 100ms or increased memory use by more than 100KB. This would let me ignore thousands of trace lines I don't need to see and would make optimisation much easier - as it is, in ZF2, the bloated framework takes 1 second just to start with the composer autoloader on our enterprise project, which results in thousands of lines I really have no use for. Somewhere along the line I do find the bigger jumps in execution time, but not after a long bout of scrolling.

Is there a default option to disable logging of "faster than X" executions, or if not, do you know of a helper shell/python script that could grep just the slower ones out?

For clarification, I am talking about the contents of the .xt file I get by running xdebug_start_trace() in my app.

Swader
  • 11,387
  • 14
  • 50
  • 84

1 Answers1

1

I know nothing about such options, but what I may suggest is to use profile instead of trace.

Here is an article how you can use it. If short, place these lines to your php.ini file:

xdebug.profiler_enable = 0
xdebug.profiler_enable_trigger = 1
xdebug.profiler_output_dir="c:\Projects"

and when you want to start profiler, run url with query parameter ?XDEBUG_PROFILE=1

This will produce a file with name like cachegrind.out.* and place it into profiler_output_dir.

That file could be viewed with CacheGrind viewer for your OS. Link above has a list of apps to view those files for different platforms. I were using wincachegrind (for Windows) to profile ZendFramework app. Very useful tool, as for me. And interface allow to see call tree, execution time, number of calls etc. Well, but I see no option to measure memory usage with it.

Viktor S.
  • 12,736
  • 1
  • 27
  • 52
  • Thanks, I just tried analyzing the cachegring with webgrind, but those results are useless to me. The values "total call cost" and "total inclusive cost" are completely irrelevant, because I can't see the flow and can't identify the big jumps and roadbumps - for instance, function1() might take 0.003s in one context, and 1.3s in another - these traces do not reflect this. Hence why I'd like to filter the trace output by performance jumps. – Swader Feb 13 '13 at 08:58
  • 1
    Hm. Not sure I completely understand what do you mean, but in wincachegrind there is show overall menu item if you right click on function. In overall panel it shows all calls of that function with time of each function, caller function and file. Besides, I've looked around the web and there are trace output visualizers written in PHP. Never used those, but possibly you will find them helpful. – Viktor S. Feb 13 '13 at 09:06
  • 1
    [Here](http://stackoverflow.com/q/1456395/1558311) is a question on SO where I saw few links to trace output viewer. – Viktor S. Feb 13 '13 at 09:08
  • I'm currently on a Mac, and there's next to no viable solutions. There's a $30 utility called MacCallGrind, QCacheGrind which I cannot seem to install, and webgrind which only gives me these values. I guess it's for lack of tools.. I'll give those php trace analysers a go - sounds like what I'm looking for. I tried webgrind which that answer mentions, will try the GUI next. – Swader Feb 13 '13 at 09:09
  • The GUI project seems promising but is very badly coded. It runs out of memory because it attempts to load the entire file and has some dependencies on the trace format and whatnot. I'll see if I can fix it and post back - otherwise, it's exactly what I'm looking for. – Swader Feb 13 '13 at 09:32
  • Try to increase memory limit in php. Something like `memory_limit = 64M` in php.ini should solve a problem with memory. – Viktor S. Feb 13 '13 at 09:36