[PATCH] D56959: [AST] NFC: Introduce new class GenericSelectionExpr::Association

[Bruno Ricci via Phabricator via cfe-commits]

riccibruno added inline comments.


================
Comment at: include/clang/AST/Expr.h:5068
 
+  Association getAssociation(unsigned I) const {
+    return Association(cast<Expr>(SubExprs[END_EXPR + I]), AssocTypes[I],
----------------
riccibruno wrote:
> aaron.ballman wrote:
> > steveire wrote:
> > > aaron.ballman wrote:
> > > > steveire wrote:
> > > > > aaron.ballman wrote:
> > > > > > Rather than gin these objects up on demand every time they're needed, I'd prefer to see the class store `Association` objects directly. I don't think it will be easy to do that and still support `getAssocExprs()` and `getAssocTypeSourceInfos()`, so I think those APIs should be removed in favor of this one. There's currently not many uses of `getAssocExprs()` or `getAssocTypeSourceInfos()` (I spot one each in Clang) so migration to the new API should not be onerous.
> > > > > > 
> > > > > > This should also have a range-based accessor version so that users aren't required to use iterative loops to access the information (a lot of the places you're already touching could use that range-based interface).
> > > > > I would prefer that too, but it doesn't seem to be possible. This is a sub-range of the `SubExprs` returned from `children()`. 
> > > > > 
> > > > > In theory, that could be split into two members, but then you would need a range library to recombine them and implement `children()`: https://godbolt.org/z/ZVamdC
> > > > > 
> > > > > This seems to be the best approach for now, and AFAIK it excludes a range-accessor.
> > > > > I would prefer that too, but it doesn't seem to be possible. This is a sub-range of the SubExprs returned from children().
> > > > 
> > > > Ugh, you're right. :-(
> > > > 
> > > > > In theory, that could be split into two members, but then you would need a range library to recombine them and implement children(): https://godbolt.org/z/ZVamdC
> > > > 
> > > > We have zip iterators that could be used to implement this, I believe. (see STLExtras.h)
> > > > 
> > > > Alternatively, we could tail-allocate the Association objects with (perhaps references to) pointers back into the Expr tail-allocated array. Not ideal, but does provide a clean interface.
> > > > 
> > > > @riccibruno may have other ideas on how to pack the arrays, as he's done a lot of this work recently.
> > > > We have zip iterators that could be used to implement this, I believe.
> > > 
> > > You're right, there is a `concat` there, but on second thought - because Association and Stmt don't share a base, I don't think it can work.
> > > 
> > > > Alternatively, we could tail-allocate the Association objects with (perhaps references to) pointers back into the Expr tail-allocated array. Not ideal, but does provide a clean interface.
> > > 
> > > Perhaps this can work, but I don't know how to do it. If you have scope for it in your part of the efforts, it would be a good way to get this unblocked.
> > > Perhaps this can work, but I don't know how to do it. If you have scope for it in your part of the efforts, it would be a good way to get this unblocked.
> > 
> > I'll put some time into it today and see where it goes. You may be right that this is more work than it's worth, so we'll see.
> I don't see what would prevent tail-allocating the array of sub-expression and the array of `TypeSourceInfo`, and removing the `getAssocExpr`, `getAssocTypeSourceInfo`, `getAssocType` interface in favor of a single `getAssociation`. Then create a range version of `getAssociation` using the fact that if you know where you are in the sub-expression array, you know where is the corresponding `TypeSourceInfo`. To know which index correspond to the selected sub-expression you could use one of the low bits in the `TypeSourceInfo` pointers.
> 
> This means that `children` is still simple to implement, and users can use a single unified
> interface via `getAssociation` and the range version `associations()`. From the point of view of the users it would be like if the `Association` objects were stored contiguously.
Made a patch which roughly the above idea: D57098


Repository:
  rC Clang

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D56959/new/

https://reviews.llvm.org/D56959