On Monday, 6 January 2014 at 16:49:44 UTC, ilya-stromberg wrote:
Unfortunately, Deimos does not good enough.
I used `libfcgi` ~2 years ago, and it was completely broken
(segmentation fault after 1-st GC memory free).
So, I agree with Jacob: we should close down Deimos. As
alternative, we should use the same review process as for
Phobos.
So there is a single library that is not "good enough". Where is
your bug report? Your pull request?
I just had a look at the libfcgi repository, and it seems like
Jonathan recommends building the binding with "ldc2 -shared" in
the readme. There are two things wrong with this:
1) -shared is not yet supported in LDC for D2 (it will lead to
GC-related crashes), and Jonathan knows this.
2) Deimos headers should *never* require actually building
something, they should be "header-only", in C terms.
So, yes, there are apparently problems with getting the Deimos
idea across (and Walter's code review practices), but I don't see
how this justifies ditching the whole idea.
The idea behind Deimos is that there should be a single of plain,
"no-frills" C bindings, because it makes exactly zero sense to
duplicate work here. Other people can build on these for
higher-level libraries. Whether these are managed in one central
place or in separate repositories doesn't matter in the end; the
thing that counts is that we have a common understanding of how
bindings that are "officially" accepted are supposed look like.
If every single C binding on code.dlang.org follows a different
naming scheme, loading convention, …, just using a C library will
become a lot less of a plug-and-play experience than it could be.
Yes, there are currently issues with the way Deimos is handled,
starting with the fact that a ridiculously small number of people
actually has push access to them (e.g. I have access to all the
D-P-L ones, but not to Deimos). In fact, I think it might even
make sense idea to give the original creator of a binding write
access to the repository after the initial review is complete,
which ensures that the author is familiar with the Deimos
conventions. There isn't really a lot to get wrong with C
bindings that would necessitate much review afterwards.
Also, we need to improve the documentation about the Deimos
standards and process. But still, I don't think we should
outright ditch the idea at this point; the situation certainly
won't get better without Deimos being in the picture.
David
