Re: [PATCH 2/2] pretty: add merge and exclude options to %(describe)

From: René Scharfe. <l.s.r@web.de>
Date: Mar 6, 2021, 16:18 UTC
Message-ID: <a28592c5-4a70-1ea8-fd73-959e28b4278d@web.de>
In-Reply-To: <30f78fce-fe1a-0a50-fa8c-e6fe14a03bf3@web.de>

Am 02.03.21 um 17:00 schrieb René Scharfe.:

Show 13 quoted lines

> Am 28.02.21 um 16:41 schrieb Ævar Arnfjörð Bjarmason:
>> Instead of taking the trouble of putting a limit in the
>> pretty_print_context so we don't call it N times for the same commit,
>> why not just put the strbuf with the result in that same struct?
>>
>> Then you can have it millions of times, and it won't be any more
>> expensive than the other existing %(format) specifiers (actually cheaper
>> than most).
>
> Each %(describe) placeholder can have a unique set of match or exclude
> arguments.  Caching them all would increase the strength of a DoS
> attack.  Caching a few of them would be OK, but is ineffective in
> reducing the strength of the attack.

The script at the bottom creates archives that illustrate the issue. On a repo generated with parameter 10 (10 files with 10 $Format:...$ with 10 %(describe) placeholders, so 1000 total), I get the following number with v2.30.1, which ignores %(describe):

Benchmark #1: git archive HEAD
  Time (mean ± σ):       2.2 ms ±   0.2 ms    [User: 0.9 ms, System: 1.0 ms]
  Range (min … max):     1.8 ms …   2.8 ms    705 runs

  Warning: Command took less than 5 ms to complete. Results might be inaccurate.

The version in next expands all placeholders and takes three orders of magnitude longer:

Benchmark #1: git archive HEAD
  Time (mean ± σ):      2.300 s ±  0.003 s    [User: 819.0 ms, System: 1200.0 ms]
  Range (min … max):    2.293 s …  2.305 s    10 runs

The proposed patch to expand only a single placeholder gets the runtime back under control:

Benchmark #1: git archive HEAD
  Time (mean ± σ):       4.7 ms ±   0.3 ms    [User: 1.8 ms, System: 2.2 ms]
  Range (min … max):     4.2 ms …   7.0 ms    451 runs

  Warning: Command took less than 5 ms to complete. Results might be inaccurate.

Using parameter 100 takes about a second to create the repo, but the git archive version in next already needs longer to tar it up than I'm willing to wait.

#!/bin/sh
n=$1
mkdir $n
cd $n
git init
for i in $(seq $n)
do
	awk -v i=$i -v n=$n 'END {
		for (j = 0; j < n; j++) {
			print "$Format:"
			for (k = 0; k < n; k++) {
				print "%(describe:exclude=x-" i "-" j "-" k ")"
			}
			print "$"
		}
	}' </dev/null >"file$i"
done
git add file*
echo "file* export-subst" >.gitattributes
git add .gitattributes
git commit -m initial
for tagno in $(seq $n)
do
	git tag -m "$tagno" "tag$tagno"
done