|Posted by Martin Nowak||PermalinkReply|
- signature.asc (OpenPGP digital signature)
- text/plain part
Welcome to the 10th sprint planning.
I'm back from vacation and currently catch up github etc.
Apparently there wasn't that much activity during the past 2-3 weeks.
If I missed something important, please let me know.
# Nightly builds (https://builds.dawg.eu/)
The nightly build service has proved reliable and it's time to move it
to official dlang subdomain or even better integrate it with
downloads.dlang.org. I still want something nicer than nginx's autoindex
(https://builds.dawg.eu/), so that would be a good opportunity to
renovate downloads.dlang.org as well. Any suggestions or volunteers welcome.
I would also appreciate any help in polishing the build script output. https://builds.dawg.eu/dmd-nightly/build.html https://github.com/D-Programming-Language/installer/tree/master/create_dmd_release
# Install script
Same goes for the install.sh script, works reliably and makes it really
quick and easy to install or switch compilers.
There has been less review and testing than I'd hoped for, but let's
still try to move on with this.
# Open tasks
I'm a bit behind and still work on the GC spinlock and dmd's slist_reset.
# ld.gold regression
Still not fixed (after 3 weeks :o), needs to be addressed ASAP.
# @nogc tracehandler
We currently workaround the issue by suppressing traces when throwing
from the GC. Overall this story is quite big b/c we need to rewrite the
complete demangle code and also a big part of the new ELF/DWARF-aware
tracehandler, so it will have to wait some more.
The implementation of the new tracehandler is somewhat simplistic
(linearly interprets dwarf_line, doesn't use debug_aranges for fast
function lookups). So I'm wondering whether we can leverage libbfd or
addr2line instead of implementing everything ourselves (and increasing
every D binary size). Also the binutils implementation should support
# releasing 2.070
I think we should delay the next release for 2 weeks (targeting Jan 14th), b/c we're already behind and a beta during the holidays doesn't make much sense.
What's the state of the new EH implementation btw? It seems this feature
was once again merged incrementally, against better knowledge.
It's extremely difficult to assess someone's work if it comes in little
slices. And it easily happens that necessary tasks are left unresolved
before the next release.
Please use feature branches to develop and complete a whole feature
before merging it. Reviews on PRs can be done incrementally w/o a need
to merge the interim states, and github makes it easy to append new
commits to an already reviewed PR.