Ary Borenszweig wrote:
Frits van Bommel escribió:
Ary Borenszweig wrote:
Now that D2 is being developed, I don't know how much time it will
take until it is finished. Once it is, what will happen to D1? Will
anyone still use it?
I think I might keep using it for a while yet. D2 has some features I
dislike. On the other hand, some others are very tempting...
I ask this because from time to time I like to add new features to
Descent to improve D1 support. Semantic analysis is pretty done for
D1, except for some bugs that are hard to fix, or at least take a lot
of time. :-P
Well, some places I'm still really missing semantic analysis in Descent:
* Nested functions (their bodies don't seem to get analyzed)
Do you have an example? It seems to work for me. Maybe you are talking
about templates? These are not analyzed.
No, I'm talking about nested functions. However, trying to create a
small example shows that they do indeed work. Unfortunately, that's
definitely *not* the case in some of the files in a project I actually
use Descent for :(.
I'll see if I can make some time later to reduce that project to a
reasonable testcase.
* Mixins (functions defined by (string) mixins don't get highlighting,
don't show up in code completion, etc.)
Again, it works for me, but this time not always (because the semantic
analysis ported from DMD isn't finished/has bugs). But I tried this
example:
[snip]
Again, this works for the small testcase but not in that same project I
mentioned above :(.
(Those mixins also use templates to generate the strings, but again in a
small testcase those very same functions actually do seem to get
recognized...)
Or are those the "hard to fix"/"take a lot of time" cases? (Or fixed
in trunk but not up on the update site?)
Semantic analysis bugs are hard to fix (and, well, kinda boring),
specially since Descent doesn't use DMD's code unmodified because I had
to introduce some optimizations and lazy loading. Template semantic
analysis is also hard, because I'd had to write it myself almost from
scratch. DMD analyzes templates in instantiations only. Maybe I could
analyze every template with a dummy instantation; the problem is I don't
know which concrete arguments to use (they might lead to static ifs
failing, etc.).
I can imagine the problems with semantic analysis of uninstantiated
templates. But that's not where my problems were.
