Re^4: Hypothesis: some magic string optimalization in perl kills my server from time to time

[leszekdubiel]

Here is the file -- 200 kilobytes (... small one):

download from Poland

download from we transfer

There is minimal test and description. Below is how I run tests. See -- one is half a seconds, and other is 20 seconds.

# ./show_me_the_problem.sh 


****************************
here are the two programs:
-rwxr-xr-x 1 root root 177 wrz 30 17:57 run_fast.pl
-rwxr-xr-x 1 root root 190 wrz 30 17:57 run_slow.pl


****************************
the difference is: 
10c10,11
< my $lst = list_extr(\( "(" . $inp . ")" )); 
---
> my $par = "(" . $inp . ")"; 
> my $lst = list_extr(\$par); 


****************************
data is the same... let's run them...
time ./run_fast.pl | wc -l
9506

real    0m0.476s
user    0m0.456s
sys    0m0.016s


****************************
time ./run_slow.pl | wc -l
9506

real    0m22.591s
user    0m22.556s
sys    0m0.012s


****************************
seee... programs are almost the same... but their time is totally dife
+erent

[dave_the_m]

I removed the no autovivification line from libr_data as I haven't got a perl to hand with that module installed, and both programs ran fast for me for all perls from 5.10.0 through 5.24.0. So that might be a thing for you to investigate first.

Dave.

[BrowserUk]

Ditto. I couldn't reproduce the OP's timing differences, but I had also commented out no autovivification as I didn't have it (I cannot think why anyone would want it), and it didn't seem to stop the code from working.

[leszekdubiel]

Ok! Thank you for checking. This has lead me to figure out that the problem exists only on currently stable distribution of Debian, while on newer software it runs as fast as expected. This is not a problem of autovivification. Tested on four of Debian installations. I will dig it deeper and maybe file a bug.

[leszekdubiel]

Here is bug report: bug