Unittesting libraries

[simendsjo]

On 10.08.2011 12:11, Jonathan M Davis wrote:
> On Wednesday, August 10, 2011 12:01:00 simendsjo wrote:
>> On 10.08.2011 09:29, Jonathan M Davis wrote:
>>> On Wednesday, August 10, 2011 09:11:53 simendsjo wrote:
>>>> On 08.08.2011 14:11, simendsjo wrote:
>>>>> Is it possible to run unittests in libraries?
>>>>> The following doesn't work:
>>>>>
>>>>> l.d
>>>>> ===
>>>>> module l;
>>>>> import std.stdio;
>>>>> int f() { return 1; } // just to make sure it's actually compiled in
>>>>> unittest {
>>>>> writeln("Unittest from lib");
>>>>> assert(false);
>>>>> }
>>>>>
>>>>>
>>>>> t.d
>>>>> ===
>>>>> import l;
>>>>> import std.stdio;
>>>>> void main() {
>>>>> writeln(f());
>>>>> }
>>>>>
>>>>>    >   dmd -unittest -lib l
>>>>>    >   dmd -unittest t l.lib
>>>>>    >   t.exe
>>>>>
>>>>> 1
>>>>
>>>> Anyone? Now I have to compile every source file from the library into
>>>> t.d instead of compiling in the library.
>>>
>>> I don't know exactly what the deal with unit tests and libraries is, but
>>> it wouldn't surprise me at all if you have to compile a module into a
>>> binary to run its unit tests. Phobos (on Linux at least) is set up to
>>> build each module separately to run its unit tests. I can see why you'd
>>> want to do this, so it may be worth creating a bug report or
>>> enhancement request for it, but I have no idea whether it's supposed to
>>> work or not.
>>>
>>> - Jonathan M Davis
>>
>> I haven't tested, but does this mean invariants and contracts is not run
>> too? In that case I think it should be fixed (if I'm not doing anything
>> wrong).
>
> They definitely should be. There's a big difference between running the unit
> tests and compiling in invariants and contracts. It could be that having built
> your library with -unittest, all of the unit tests are compiled in. They're
> just not run, because that requires some instrumenting from the runtime to run
> them before main. If it doesn't put the unit tests from the library in the
> list, then they won't be run. However, invariants and contracts should be
> compiled directly into the code such that there should be no need for separate
> code to call them like there is with the unit tests.
>
> Regardless, not running the unit tests from a library may be by design, but it
> is _definitely_ a bug if the invariants or contracts aren't run. But since
> people have hit invariants in Phobos when running their own programs (though
> that requires Phobos to have been compiled without -release), I'd be very
> surprised if it invariants or contracts were broken with regard to libraries.
>
> - Jonathan M Davis

I added it to druntime as you think that's where the problem is.
We'll see what Sean or others say about it.
http://d.puremagic.com/issues/show_bug.cgi?id=6464