Benchmark python specfile a bit

[praiskup]

We are going to process all spec files with python-specfile now. We already count with parse errors, though parsing might be really expensivefor some packages. We should test it on texlive.spec e.g.

[praiskup]

@xsuchy claims it is non-problem

[praiskup]

from specfile import Specfile
import time

specfile_path = "texlive.spec"

script_start = start = time.time()

def benchmark(message):
    global start
    now = time.time()
    print(f"{message}: {now-start} seconds")
    start = now


spec = Specfile(specfile_path)
benchmark("loaded")

print(spec.expanded_name)
benchmark("name parsed")
print(spec.expanded_epoch)
benchmark("epoch parsed")
print(spec.expanded_version)
benchmark("version parsed")
print(spec.expanded_release)
benchmark("release parsed")

print(f"The script took {time.time() - script_start} seconds")

Running the script, it hangs for a few minutes here:

loaded: 15.034465312957764 seconds
Getting name

[praiskup]

Continues:

name parsed: 1852.5356392860413 seconds
Getting epoch

[praiskup]

Memory consumption 827M, seems acceptable for the builder, not sure about backend.

[praiskup]

I accidentally canceled the terminal, but the epoch was parsed after another ~1750s. On i7-10850H. @nforro, isn't there a trick to do this faster than in say 2 minutes?

[nforro]

Sorry, this is crazy. Let me try to do some optimizations.

[nforro]

With packit/specfile#240, on my laptop:

loaded: 4.407037734985352 seconds
texlive
name parsed: 6.255786657333374 seconds
11
epoch parsed: 10.084052324295044 seconds
2023
version parsed: 9.657040357589722 seconds
69
release parsed: 16.89114236831665 seconds
The script took 47.29507040977478 seconds

To make it faster, you can do this (to parse tags only once):

spec = Specfile(specfile_path)
benchmark("loaded")

tags = spec.tags().content
print(tags.name.expanded_value)
benchmark("name parsed")
print(tags.epoch.expanded_value)
benchmark("epoch parsed")
print(tags.version.expanded_value)
benchmark("version parsed")
print(tags.release.expanded_value)
benchmark("release parsed")

print(f"The script took {time.time() - script_start} seconds")

Result:

loaded: 5.885213136672974 seconds
texlive
name parsed: 9.854274272918701 seconds
11
epoch parsed: 0.0469050407409668 seconds
2023
version parsed: 0.04676461219787598 seconds
69.fc37
release parsed: 0.05761575698852539 seconds
The script took 15.89078426361084 seconds

[nforro]

Oh, and to get release without the dist tag, you can do:

print(spec.expand(tags.release.value, extra_macros=[("dist", "")]))

But it will slow things down a bit again:

loaded: 4.443751811981201 seconds
texlive
name parsed: 8.05765676498413 seconds
11
epoch parsed: 0.04661202430725098 seconds
2023
version parsed: 0.04650592803955078 seconds
69
release parsed: 7.663807153701782 seconds
The script took 20.258345365524292 seconds

[praiskup]

Nice, thank you for the fix, @nforro! This is likely one of the worst scenarios, and if we can parse such a spec file in less than a minute, it is more than good.

[praiskup]

Once in stable, we should Require: python3-specfile >= 0.19.0.

[praiskup]

https://bodhi.fedoraproject.org/updates/?search=python-specfile

[praiskup]

Jakub promised to specify the specific version in our specfiles once it is in stable (changed assignee).

[praiskup]

#2800 fixes this