Idea for a new tool

[Mathias Lang via Digitalmars-d]

On Friday, 14 October 2016 at 13:21:18 UTC, Andrei Alexandrescu 
wrote:
> On 10/14/2016 09:20 AM, rikki cattermole wrote:
>> On 15/10/2016 2:17 AM, Andrei Alexandrescu wrote:
>>> On 10/14/2016 09:15 AM, rikki cattermole wrote:
>>>> On 15/10/2016 2:13 AM, Andrei Alexandrescu wrote:
>>>>> https://issues.dlang.org/show_bug.cgi?id=5051 -- Andrei
>>>>
>>>> As asked for time and time again, this shouldn't be an 
>>>> external tool
>>>> handling it but dmd-fe should.
>>>
>>> What is dmd-fe?
>>
>> dmd front end
>
> Yah, making the front end a library would probably be a good 
> way to create such a tool. -- Andrei

And to enable the creation of many others.

I've looked into pluging DMD-FE inside a Vibe.d server a while 
ago, in order to be able to link usages to declarations (and the 
other way around) using a Github API.

There are a couple of things that prevented any usage of the FE 
as a library:
- The frontend uses loads of globals and static - not only does 
it makes for an horrible library interface, but it makes it so 
that you cannot just discard the result of a semantic analysis 
and start over.

- DMD currently disables the GC. Enabling it causes random errors 
during semantic.

- All files in DMD should have to be moved to match their package 
definition ( explanation in 
https://issues.dlang.org/show_bug.cgi?id=16071 , there was also a 
thread on this topic a couple of weeks ago).

- The frontend depends on the backend when inline ASM is involved 
(and if you want to parse druntime's modules...). You can make 
this a soft dependency in some cases, but then the frontend just 
won't see your `ret` for example, so your flow control should not 
depend on your inline ASM.

- There were also random errors, most likely due to memory 
corruption, caused when compiling the frontend in (without any 
actual usage). For example, a global AA in Vibe.d got it's length 
reset.


While those issues are fixable with some effort, there doesn't 
seem to be any particular interest upstream towards making this 
work. If it is indeed a desirable direction, then maybe we could 
head to dmd-internals about what needs to be done, and how.