org/eolang/lints/Program.defect Performance: Slow Execution

[volodya-lombrozo]

I have added the usage of Program.defect to the jeo-maven-plugin, and the plugin now runs three times slower than usual. I conducted performance profiling and found that the org/eolang/lints/Program.defect method accounts for 87% of the total execution time.

Please address the performance issues related to org/eolang/lints/Program.defect.

flamegraph-disassemble-28078.html.txt

[volodya-lombrozo]

@yegor256 Take a look, please. It's a blocker as well.

[yegor256]

@h1alexbel maybe you can help us investigate this?

[volodya-lombrozo]

@yegor256 Please note, this is an urgent issue. It actually blocks any new features in jeo-maven-plugin. We need to resolve it as soon as possible.

[yegor256]

@volodya-lombrozo I'm afraid, we can't solve it entirely. Some extra performance burden is expected and inevitable. Earlier, we had all linters inside eo-maven-plugin. Now they are here. Obviously, they take time. In your particular case (you may have very large XML files), the performance cost may be very high. The best you can do immediately is simply stop checking all XMIR files via linters, but only check some small files. You need linting simply to validate the correctness of your output -- no need to check all big files, just check the smallest one.

[volodya-lombrozo]

@yegor256 It's just faster to fix the performance issue here. Otherwise, I will need to 'invent' a lot of workarounds to hide the original problem, which I will need to remove as soon as a proper solution is implemented.

[yegor256]

@volodya-lombrozo I'm not sure there is any "issue" here: there are just 30+ lints which take time when XML files are huge. It's not an issue, it's cost of doing linting :)

[h1alexbel]

@yegor256 @volodya-lombrozo how about we add new benchmark test to check Program.defect performance?

[yegor256]

@h1alexbel not a bad idea to create a benchmark: it should calculate how much it takes to lint a small XMIR, a large, an X-large, an XXL, etc. We can publish this information in README, similar to how we publish it here: https://github.com/objectionary/hone-maven-plugin/blob/master/.github/workflows/benchmark.yml

[h1alexbel]

@yegor256 assign me, I will try to

[yegor256]

@h1alexbel take it, in #134

[volodya-lombrozo]

I've provided one more profiling round for the version 0.0.27. Here are results:
flamegraph-disassemble-56062.html.txt

As we can see, the objectionary/lints still takes unreasonable amount of time