Bug#732836: ben: tracker could do with a performance enhancement

[Niels Thykier <niels@thykier.net>]

On 2013-12-22 13:15, Stéphane Glondu wrote:
> Le 22/12/2013 13:03, Niels Thykier a écrit :
>>>> In particular, it seems to be very slow for ben files containing a lot
>>>> of "OR" relations; even if these are exact package names (see attached
>>>> file).
>>>
>>> Did you try:
>>>
>>>   .depends ~ "foo|bar"
>>>
>>> instead of:
>>>
>>>   .depends ~ "foo" | .depends ~ "bar"
> 
> Sorry, I actually meant ~ /foo|bar/
> 

Ah, okay.  Though it triggers a "Pcre.Error" with the rewritten
auto-linux file:

"""
Generating (ongoing) auto-linux
E: Pcre.Error(_)
"""

I suspect that the underlying regex engine does not like that enormous
expression.  Besides that, it does seem to work and it takes only 3
minutes (excluding the above mentioned file for obvious reasons).

If you want, I can send you the resulting file that ben/Pcre doesn't like.

> The code for ~ "foo" does indeed look suboptimal (to say the least)...
> 

:)

>> Thanks for the tip; it does seem to improve the general performance of
>> these files considerably.  Now all 40 files on release.d.o are processed
>> in 3m and 30 seconds[1].  Previously it was 7 minutes excluding the one
>> I attached (which seemed to take more than 20 minutes alone).
> 
> Is the result right with double quotes?
> 

By the looks of it; no.


>> It does not seem like the "|" inside a string is documented on ben.d.n[2].
> 
> It is captured by <regex>.
> 
> 
> Cheers,
> 

:)

~Niels