On Saturday, 7 May 2022 at 20:14:51 UTC, Witold Baryluk wrote:
Good you mentioned that! I was not aware of the sponsorship
program, and now that I know, I gladly will chip in (well, just
did it moments ago). For testing infrastructure, I would
suggest tracking compilation speed and memory usage and output
binary size of GDC on amd64 and aarch64 at least (to detect
compiler getting slower, or due to growth of Phobos /
druntime), and having a public website showing this data.
Something like this maybe https://fast.vlang.io/
Cheers.
Thanks for the suggestion. Vladimir did in fact do that for dmd
last decade, it only ran for a couple years though. Have been
thinking about maybe reviving it every so often.
https://blog.cy.md/2015/05/05/is-d-slim-yet/
Adam (maybe in a TWID post) did a few months back lament that D1
vs D2 equivalent code compiles slower with the latter. The bulk
of which in the trivial case came from Druntime and how many
modules are imported (D1 object.d had no imports, D2 object.d
imports around 25 modules). Remove the excessive imports and the
original speed was observed again. That's only one small example
though of where perceived slowness comes from the library
becoming more complex over time - and I expect it only to
increase as more of the old opaque compiler-library interface is
replaced with a templated interface that exposes the guts of what
each helper does (for improved run-time performance, of course).
