Re: [Haskell-cafe] GDB and ticks

I've just done some non-scientific benchmarks that show that in code with a lot of (I presume) unsafe FFI calls such as buffered `putStr` on Linux, performance can drop by half and even more with `+RTS -V0`. I can share if anybody is interested. BTW, this is orthogonal to `-threaded` which in my benchmarks slows no-FFI code with a few threads by 20%, safe FFI calls-heavy code (SDL2 graphics) by 50% and speeds up unsafe FFI-heavy code (`putStr`). On Mon, Oct 18, 2021 at 4:35 PM Brandon Allbery wrote:

Sadly, it also comes with a price: everything including GCs runs at every context switch instead of waiting for an appropriate number of ticks to pass. (That said, I've also been bitten by the interrupt issue.)

On Mon, Oct 18, 2021 at 10:28 AM Donn Cave wrote:

...

quoth Henning Thielemann

...

On Mon, 18 Oct 2021, Sylvain Henry wrote:

...

The timer isn't only used for profiling, it is used for scheduling too. You can disable it with `+RTS -V0` (cf https://ghc.gitlab.haskell.org/ghc/doc/users_guide/debug-info.html#tutorial)

Cool, this works!

What does it break? Just about anything I'd run that was written in Haskell must run this way, to avoid fatal interrupts in things like socket I/O, so it's essentially a default - everything must be built with -rtsopts, and run with GHCRTS=-V0 environment.

Donn _______________________________________________ Haskell-Cafe mailing list To (un)subscribe, modify options or view archives go to: http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe Only members subscribed via the mailman list are allowed to post.

-- brandon s allbery kf8nh allbery.b@gmail.com _______________________________________________ Haskell-Cafe mailing list To (un)subscribe, modify options or view archives go to: http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe Only members subscribed via the mailman list are allowed to post.