configuration parser

[rtbo]

Hi guys,

Following #1832 and #2007, yet another config file pull request.
Hopefully the last one!

[dnadlinger]

You can delete the whole line, then.

[kinke]

I like it, thanks. Some unittests wouldn't hurt; I guess we'll have to switch from reading directly from file to reading the full config string then.

[rtbo]

Some unittests wouldn't hurt

I Agree. I don't think any build target exists yet to run unittest blocks in the ldc executable.

[JohanEngelen]

run unittest blocks in the ldc executable.

I am very strongly against adding unittests inside the source. It drowns code and promotes only shallow unittesting. Adding a general unittest framework to our repo would be nice, but for now perhaps you can just add some lit tests with %ldc -conf=... .

[dnadlinger]

I am very strongly against adding unittests inside the source. It drowns code and promotes only shallow unittesting. Adding a general unittest framework to our repo would be nice, but for now perhaps you can just add some lit tests with %ldc -conf=... .

A separate unittest executable and using the normal D facilities would be just fine.

[kinke]

I like the separate LDC unittest executable. As building it only recompiles the D modules, it doesn't even take a minute on our CI systems.

[kinke]

Any remaining concerns?

[JohanEngelen]

It'd be good to get this in 1.2-alpha too

[dnadlinger]

We can probably overhaul the code to make use of Phobos in the future, but it seems like everything should be okay for now.

[kinke]

It'd be good to get this in 1.2-alpha too

I don't think so. See #1960 (comment).

[dnadlinger]

I don't think so

Ideally, this wouldn't affect users at all due to compatibility with the old implementation.

[kinke]

It doesn't. But it gets rid of the libconfig runtime dependency and as such distro maintainers should get rid of it too. In order not to annoy them with changes in 1.2 and 1.3 and to accelerate the 1.2 release (didn't we say one month ago that we're going to put out an alpha/beta immediately?), I'd prefer to delay it to 1.3, whose release cycle can theoretically (very theoretically given the very limited current manpower) be started these days anyway now that merge-2.073 is green too.