449428 - dehydra: want process_decl callback

Status: RESOLVED FIXED

(Developer Infrastructure :: Source Code Analysis, defect)

Description

[Benjamin Smedberg]

I want a callback for declarations:

process_decl(decl)

this should be fired for function and variable declarations that occur outside of class scope (process_type already hooks function/var declarations that occur within a class)

Comment 1

[(dormant account)]

To clarify:
Want template decls, typedef decls, variable decls, and function (forward+implementation?) decls.

Comment 2

[Benjamin Smedberg]

templated function, templated class/struct, typedef, variables, and function decls. I don't need variable or function implementations, because those are already handled by process_var/process_function

Ideally that would include decls of specialized or partially-specialized classes, e.g.

template<class T, class U>
Foo
{ ... }; // process_decl for Foo<T, U>

template<class U>
Foo<T, nsISupports>
{ ... }; // process_decl for Foo<T, nsISupports>

Comment 3

[(dormant account)]

Attachment: expose more stuff via process_var()

The idea here is to expose stuff dehydra already sees(no mods to gcc) to process_var and rename it to process_decl. I think it would be terrible to have a process_var/process_decl split, so I'd rather rename the function.

See the testcase for what's supported:
a) function forward decls are now seen, unless decl smashing occurs
b) typedef declarations
c) template declarations. Partial specializations are listed similar to template arguments. I'm not sure how to support full specializations as I haven't found what gcc does with them yet.

Once you review this, I'll also post a rename patch for process_var

Comment 4

[Benjamin Smedberg]

Comment on attachment 332995 [details] [diff] [review]
expose more stuff via process_var()

There's an odd commented-out "else" in the testcase.

I'm happy with this patch as a first pass, but comments to follow:

Comment 5

[Benjamin Smedberg]

Template specializations don't do exactly what I want yet, but I'm happy to file a followup for them:

template<class T, int I>
class Foo
{
  T *mT;
};

template<class T>
class Foo<T, 1>
{
  T mT;
};

With your current patch, I get a single call to process_var with Foo and a .specializations array. This specializations array lists the template parameters for the specializations, but not the actual class declaration. Because the class declaration can be entirely different, I'd either like:

* the .specializations array to have the specialized type object
* or a separate call to process_decl with the specialized type.

Comment 6

[(dormant account)]

I'll dig some more. I might've missed where gcc hides this info.

Comment 7

[(dormant account)]

Attachment: this gives you template decls..apply on top of previous

So I guess I can't pull out full specializations and full partial info out of template_decl, at least it isn't obvious. On the other hand, I was able to get the data from elsewhere. 
This should work for you.

Anything else you need?

Comment 8

[Benjamin Smedberg]

applied and basic testing seems to work fine. There's a pre-existing issue with complete speicializations of templates, caused by bug 418170, that I'll file separately.

Comment 9

[(dormant account)]

pushed