On Tuesday, 13 December 2016 at 21:30:05 UTC, Andrei Alexandrescu
wrote:
On 12/13/2016 02:37 PM, kinke wrote:
Hey all,
on behalf of the LDC team I am proud to announce the new
1.1.0-beta6
release!
It's based on the 2.071.2 frontend and standard library and
supports
LLVM 3.5 up to current trunk (4.0).
This is awesome! Could you please tell what the expected lag
time is between a dmd release and an ldc release? Also,
obviously what we could do to improve that. Thanks! -- Andrei
Do you mean the time it takes for LDC master to reach DMD release
parity, or do you mean e.g. LDC 1.0.0 -> 1.1.0?
The former is dependent on merging the ddmdfe in ours and the
dealing with any conflicts in the changes we make to ddmdfe, (we
bracket these in version(IN_LLVM) and version(IN_LLVM_MSVC)
blocks to ease this). Johan is usually pretty quick, but
identifying and dealing with any regressions that arise takes
longer.
The latter depends on the feature set we intend to implement and
bugs and regressions and user feedback. For example 1.1.0-beta3
introduced a regression with dub causing all dub projects to fail
to build, we got that report but not much else because hardly
anyone was using beta3.
Someone (was it you?) suggested splitting the ddmdfe off (or was
it have everything under the dlang repo?), and have it be a
dependency for each of the backends, so that the frontend stays
in lockstep and we can identify regressions earlier, not sure how
this would impact GDC. We also maintain druntime in a similar
fashion to ddmdfe, although with a lot more additions for llvm
features, probably less worth doing but still worth considering.
This would also have the advantage of increased cross-visibility
thus reducing regression times. The same repo solution would also
increase the number of people familiar with the LDC codebase and
therefore likely to help report and fix issues.
