Rant on quality

[H. S. Teoh]

On Sun, Dec 15, 2013 at 10:51:29PM +0100, deadalnix wrote:
> As I've mentioned several time, dmd 2.064 broke a ton of tests in
> SDC. Because of some bad closure codegen.
> 
> By now, I'm pretty sure that most of these tests won't pass when DMD
> is fixed. Show must go on, and having such a blind spot, it is
> pretty much a guaranteed result.
> 
> Moreover, master do not even compile anything (or, to be fair, it
> does compile but nothing can be linked). It seems that the new
> behavior for template symbol generation is completely failing.

Really? I haven't hit that problem yet, and I've been compiling a bunch
of currently-on-the-backburner D projects with git HEAD. Have you tried
-allinst?


> The whole thing is constantly broken in various way, and I spend
> most of my dev working with known codegen bugs, or with homemade
> version of DMD. I'm willing to do that because I'm a D enthusiast,
> but realistically, most user would have quit for long in my position
> by now.

It's sad, but I find that D is still developing at a fast enough rate
that I simply have to live on git HEAD. It seems the least painful way
to keep code up-to-date with the latest languages changes. Many projects
that I haven't touched for half a year or a year ago, no longer compile.
Of course, many cases are actually good cases (the compiler now catches
bad code or poorly-written code where previously it accepted them), but
there have been a few regressions.


[...]
> Right now, I puzzled by the fact that the new template symbol
> generation is pushed forward as it is clearly not working. Not even
> remotely close, I have PAGES of linker error (it is not some tricky
> edge case that is failing here).

Obviously YMMV, but I haven't hit any serious issues with that. And
that's part of the problem with the current process: those people who
are using git HEAD haven't seen problems (the issues they did see have
been fixed), but people who don't use git HEAD have code that breaks. It
seems to be that the only way out is to have people compile their code
with git HEAD (even if they're not actually using git HEAD for
production, but just to compile it and report incompatibilities,
regressions, etc.). Otherwise too many things just don't get caught
until release time, and then suddenly there's a ton of breakages that
nobody anticipated.


> I though the priority was to improve quality. That mean pushing for
> feature when they are ready. Not when whoever feels like it.

The new template symbol generation code has been in for a while now, and
the issues that were reported have been fixed AFAIK (or maybe there's
one or two that are still open in bugzilla?). In any case, I haven't had
trouble with it, so obviously it's working for *some* people. Just not
enough people, it seems. So obviously more people need to be trying out
git HEAD and reporting failures on a more consistent basis. Otherwise,
it *seems* that things are working but some important use cases are
missed.


T

-- 
We are in class, we are supposed to be learning, we have a teacher... Is it too much that I expect him to teach me??? -- RL