@GergelyOrosz This resonates a lot. What really helped me was switching to engineering management a couple of years ago, which shifted my focus toward higher-level goals. In hindsight, that was fantastic preparation for the AI era.
@OndrejMirtes@PierreJoye@ThePHPF The world would be a different place if we would have more Ondrejs in general.
Both of you did and still do an amazing job and it's never enough to say thank to you.
@AJoshDickerson@VotrubaT If we just use agents to extract minimal repros from our false positives/negatives and open upstream issues, everyone wins.
PHPUnit & framework extensions will be needed of course, but feeding Mago real-world edge cases right now is the biggest lever we (the PHP community) have.
@AJoshDickerson@VotrubaT Raw speed is only half the battle, feature parity is the bigger blocker I would say.
Large private legacy codebases are a goldmine for this. Millions of LOC, full of edge cases where PHPStan still does a much better job.
@AJoshDickerson@VotrubaT Out of curiosity, are all 50 tweaks purely internal logic, or could some be created as pull requests upstream to Mago? Would be helpful for the ecosystem
@azjezz I'm going through some of my private repos and try to extract all of the stuff that I mean into reproducible chunks. Thanks for merging the stuff so quickly!
I'll send some more PRs during the next days.
@mattpocockuk Nice analogy! People often have trouble leaving these sessions. The skill keeps going, and they forget they can stop. You can say the questions are getting silly and you already have a clear plan.
Leave the meat on too long and it burns. Rare, medium, and well done are all fine.
@VotrubaT Very useful, we hooked it up with a PHP extension written in rust that collects real runtime data on prod and are currently in the process of eliminating false positives with custom providers.
@mattpocockuk@infection_php From my experience this is still an unknown topic for many devs and wrapping your head around that was a challenge in teams I was part of.
Unit testing or SA tools are easier: either the tool fails or it passes.
Interpreting mutation testing results need more judgement, AI helps
@mattpocockuk I use a deterministic tool for mutation testing (in my case @infection_php), but I also introduced a mutation testing advisor agent, that teaches you how to run mutation testing, but more importantly how to deal with the results ("Should I kill this mutant in the logger call?")
@brendt_gd Very interesting comparison. If both runners execute the same tests in a single process and produce the same results, what is the main source of the performance difference?
Is the yours lacking PHPUnit features? Any optimizations that could potentially benefit PHPUnit as well?