[cfe-dev] How to implement equals(0) in clang-query

[Mads Ravn via cfe-dev]

Thank you both for your input.

I think I will dig into it over the weekend. Hopefully I can make a little
headway with it :)

Best regards,
Mads Ravn

On Mon, Jan 9, 2017 at 6:02 PM Samuel Benzaquen <sbenza at google.com> wrote:

> On Mon, Dec 26, 2016 at 3:45 PM, Mads Ravn <madsravn at gmail.com> wrote:
>
> Hi,
>
> I have been wondering how to implement 'equals' for integer literals in
> clang-query. It doesn't seem to be working for string literals either. In
> Registry.cpp it says: "TODO: Here is the list of the missing matchers,
> grouped by reason." - here 'equals' is listed under 'Other'.
>
> So I guess there is a reason why it is not working for clang-query. Does
> anyone know what this reason is?
>
>
> I put it in the list.
> It requires more work than seemed appropriate at the time.
> The function is a template, which we don't support yet in the dynamic
> matcher layer.
> Also, it supports types that are not supported in the parser, like floats.
>
>
> I wouldn't mind fixing it, but the few things I have tried have all
> resulted in failures. I am not sure why it works in clang-tidy, but not in
> clang-query. Aren't they using the same code-base for matchers? I can see
> equals implemented in ASTMatchers.h, but why can't clang-query use that?
>
> Can anyone point me in the right direction as how to implement this in
> clang-query? I imagine there is some blocking issue since it is not
> implemented already.
>
>
> I think the "easiest" way is to make a custom dynamic matcher.
> But we tried very hard not to make these.
> Instead, we made the original matchers compatible to make sure we are
> running the same code in both.
>
> There are a couple ways to make this work:
>  1) Remove the template on equals() and list all the overloads we want
> manually.
>  2) Make equals a functor, add a type list of "accepted types" and add a
> special marshaller for that pattern.
>
> (2) is a lot of work, but would be a generic solution.
> This is how polymorphic matchers work.
> However, I think for this case (1) is best. I don't know if we would reuse
> the generality of (2).
>
> Another option similar to (1) is to add the specific overloads we want in
> clang-query and leave the template to handle everything else.
>
>
> Best regards,
> Mads Ravn
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20170112/93352849/attachment.html>