Do we support conditional targets in a BUILD file?...
# developmentw
wide-midnight-78598
02/20/2025, 3:46 AMDo we support conditional targets in a BUILD file? As in - inject this only if the backend is loaded? Could this be provided as part of a macro?
wide-midnight-78598
02/20/2025, 3:47 AMOr alternatively, use a macro to shim a target - with the real one conditionally replacing it?
wide-midnight-78598
02/20/2025, 3:50 AMThis commit is the difference of about 20ish seconds per ANYTHING on my machine, to around 6 seconds:
https://github.com/pantsbuild/pants/compare/main...sureshjoshi:pants:fewer-backends
JVM being the primary contributor.
If I could get JS/TS out (we seem to be using them for some reason in the repo), I think we could shave another 2 seconds off
b
broad-processor-92400
02/20/2025, 3:54 AMone could hack something up with
if env("..."): ....pants.bootstrappants.tomlf
fresh-mechanic-68429
02/20/2025, 4:23 AMThis commit is the difference of about 20ish seconds per ANYTHING on my machine, to around 6 seconds:Is this for the first command (so pantsd must start) or even after that? cold boot pantsd performance has been something I've noticed on our builds but haven't really investigated yet
w
wide-midnight-78598
02/20/2025, 4:27 AMone could hack something up withOof, yeah, just hacks all the way downor similar, but setting the env vars automatically would require some magic (e.g. haveif env("..."): ...check.pants.bootstrapor something)pants.toml
wide-midnight-78598
02/20/2025, 4:28 AMIs this for the first command (so pantsd must start) or even after that? cold boot pantsd performance has been something I've noticed on our builds but haven't really investigated yetThis is specifically in the pants main repo, because there is so much stuff that needs to be churned through on every change. Outside of the repo, I haven't had that much of an issue. Noticeable, but in the order of a second or two-ish
wide-midnight-78598
02/20/2025, 4:29 AMThere's a safe bet that the rule solver is taking up a bunch of time - but none of the quick hacks will reduce that. That's a longer process, which is tediously ongoing
f
fresh-mechanic-68429
02/20/2025, 4:41 AMInteresting, what I've seen on our macs is ~10s till first rule can run, but in our CI (linux, in docker) we're at ~27s to first rule, so its considerable.
w
wide-midnight-78598
02/20/2025, 4:43 AM🤮
wide-midnight-78598
02/20/2025, 4:43 AMLots of backends? Lots of targets?
f
fresh-mechanic-68429
02/20/2025, 4:49 AMI think our external plugin (hosted on an internal pypi feed, but its already downloaded in these timings) is the main cause, thats the only "obvious" thing I see in the debug logs. We have a good number of rules, but I don't think its that many
w
wide-midnight-78598
02/20/2025, 4:49 AMThat seems like a lot of time 🤷
You can vendor it directly into your repo, and it should be pretty quick - unless it runs on startup
4 Views