Make {system} removal faster [#1794970]

Throwaway issue to help experimenting now that #1608842: Replace list of bootstrap modules, enabled modules, enabled themes, and default theme with config is green. please dont review this is only for bot.

Comment	File	Size	Author
#4	1794970.patch	95.46 KB	chx
#4
#4	interdiff.txt	3.81 KB	chx

Support from Acquia helps fund testing for Drupal Acquia logo

Comments

Comment #1

alexpott

he/they

English

🇪🇺🌍

CreditAttribution: alexpott commented 26 September 2012 at 00:43

So here's my investigations so far:

Site 1, HEAD, Devel enabled, Test user created (not admin) viewing front page - http://new.vit-al.co.uk/xhprof/?run=50623310cc72c&source=

Site 2, HEAD + Patch, Devel enabled, Test user created (not admin) viewing front page - http://new.vit-al.co.uk/xhprof/?run=5062330a62b7e&sort=ct&source=

Site 1 makes 14,358 function calls
Site 2 makes 39,100 function calls

Now this is actually because of the devel module's devel_boot function... unfortunately because I'm using devel to generate the runs... will redo later... but issue is with the line drupal_get_filename('module', 'user', 'modules/user/user.module'); which causes all of the extra function calls... if the load of the user module is removed then the HEAD+patch uses nearly the same number of function.

Maybe simpletest is causing something similar to occur?

Comment #2

alexpott

he/they

English

🇪🇺🌍

CreditAttribution: alexpott commented 26 September 2012 at 00:45

Because what I really don't understand is why this patch has not resulted in a quicker test run!

Enabling aggregator in HEAD:
Overall Summary
Total Incl. Wall Time (microsec): 978,908 microsecs
Total Incl. CPU (microsecs): 920,947 microsecs
Total Incl. MemUse (bytes): 46,562,736 bytes
Total Incl. PeakMemUse (bytes): 46,689,968 bytes
Number of Function Calls: 131,369

Enabling aggregator in HEAD + Patch:
Overall Summary
Total Incl. Wall Time (microsec): 786,837 microsecs
Total Incl. CPU (microsecs): 752,965 microsecs
Total Incl. MemUse (bytes): 42,011,200 bytes
Total Incl. PeakMemUse (bytes): 42,131,608 bytes
Number of Function Calls: 108,001

Comment #3

chx CreditAttribution: chx commented 26 September 2012 at 05:24

Running a test from the browser IS faster. Ran the add feed test.I have grepped the xhprof output dir for prepareEnv.

With patch

Overall Summary
Total Incl. Wall Time (microsec): 16,650,810 microsecs
Total Incl. CPU (microsecs): 8,496,113 microsecs
Total Incl. MemUse (bytes): 16,562,472 bytes
Total Incl. PeakMemUse (bytes): 17,563,344 bytes
Number of Function Calls: 1,244,067

Without patch

Overall Summary
Total Incl. Wall Time (microsec): 20,277,706 microsecs
Total Incl. CPU (microsecs): 10,325,993 microsecs
Total Incl. MemUse (bytes): 30,384,992 bytes
Total Incl. PeakMemUse (bytes): 31,672,768 bytes
Number of Function Calls: 1,257,827

The patch beats HEAD summarily. Conclusion: the run test script must do something nasty.

Comment #4

chx CreditAttribution: chx commented 26 September 2012 at 06:47

File	Size
interdiff.txt	3.81 KB
1794970.patch	95.46 KB

So, the problem I think is that the run once part of the script runner calls simpletest_classloader_register which does a rebuild on every test. Untrivial to fix... but here's something.