Improve perfomance of UUID.hex and UUID.__str__ by using bytes.hex()

[booqoffsky]

Feature or enhancement

Proposal:

In my tests, using bytes.hex() speeds up calling the UUID.hex property.

Additionally, using the hex property and f-strings also provides some speedup in calling uuid.UUID.str.

import uuid

uuids = [uuid.uuid4() for _ in range(100000)]

t1 = time.monotonic()
for u in uuids:
    u.hex
    # str(u)
t2 = time.monotonic()

print(t2 - t1)

Results before and after the fix:

hex
before: 0.021755493999989994
after:  0.01465080400066654 

str
before: 0.06381790500017814
after:  0.05134949700004654

Has this already been discussed elsewhere?

No response given

Links to previous discussion of this feature:

No response

Linked PRs

• gh-131196: Improve perfomance of UUID.hex and UUID.__str__ by using bytes.hex() #131197

[picnixz]

In other words, it boils down between whether '%032x' % self.int is faster or not than int->bytes conversion + bytes.hex(). So to really know whether it's faster or not, I would instead check the performances of "%032x" % self.int vs x = self.int.to_bytes(16); x.hex().

Now, I would appreciate benchmarks using timeit or pyperf instead of manual benchmarks. They are a bit more precise and stable (pyperf also shows the standard deviation which is kind of good to know). While it's a noticeable difference overall, we're actually only gaining 10^-8 for each call (since you're doing lots of objects). Finally, are the benchmarks performed on a DEBUG build, release build or optimized (PGO or PGO+LTO) build?

[booqoffsky]

Thanks for the answer!

This is how the interpreter was built:

./configure --enable-optimizations --with-lto
make -s -j8

Here are the measurement results:

nice -n 19 taskset -c 4 python3 -m pyperf timeit --affinity=4 --min-time 1 --processes 1 -s 'import uuid; u = uuid.uuid4()' '"%032x" % u.int'
# Mean +- std dev: 70.9 ns +- 0.2 ns

nice -n 19 taskset -c 4 python3 -m pyperf timeit --affinity=4 --min-time 1 --processes 1 -s 'import uuid; u = uuid.uuid4()' 'x = u.int.to_bytes(16); x.hex()'
# Mean +- std dev: 61.1 ns +- 0.3 ns

[picnixz]

Mmh. The improvement is a small, but it's still a improvement I guesss. It's still > 10% improvement!

[sobolevn]

Thank you for the PR! Please, feel free to suggest more optimizations.