Alias more when strengthening #653
Conversation
|
This patch reduces the produced HTML in accessor_core from 1.8 megs to 570k |
jonludlam
force-pushed
the
alias-all-when-strengthening
branch
from
02220e9 to
d030c77
Compare
|
This needs some extra work - it's leading to broken links. For example, the first line in |
jonludlam
force-pushed
the
alias-all-when-strengthening
branch
3 times, most recently
from
4e6a198 to
e5e7b23
Compare
Not just the ones that weren't already aliases
The logic behind stripping the Alias was that in the vast majority of cases, the canonical path points to a module that's the alias of the actual hidden module. This isn't always a valid assumption - for example in `base.caml`, which creates canonical aliases to stdlib.
The previous version would resolve the canonical path, then try to move as much of it into the identifier as possible. For example, if a module in, say, `Base.String` was referencing a type `t` in `Base.Buffer`, it would resolve `Base.Buffer.t` (the canonical path), then see if `Buffer.t` was the same thing, then check just `t`, where in these cases, `Base`, then `Buffer`, then `t` are fully qualified but not fully rendered identifiers. Unfortunately if the canonical path was not expanded we'd end up with the wrong path rendered. For example, `Caml.Buffer` is an alias to `Stdlib.Buffer`, but it is marked as canonical. We would then end up with `Stdlib.Buffer` where we would have liked `Caml.Buffer` or even just `Buffer` to have appeared.
jonludlam
force-pushed
the
alias-all-when-strengthening
branch
from
e5e7b23 to
f19caea
Compare
src/xref2/strengthen.ml
Outdated
| | None -> (item :: items, s) | ||
| | Some m' -> (Module (id, r, put (fun () -> m')) :: items, id :: s)) | ||
| let m' = module_ ?canonical (`Dot (prefix, name)) (get m) in | ||
| (Module (id, r, put (fun () -> m')) :: items, id :: s) |
There was a problem hiding this comment.
The whole computation of m' could be delayed (it's not used otherwise)
| (fun s mid -> Subst.path_invalidate_module (mid :> Ident.path_module) s) | ||
| Subst.identity strengthened_modules | ||
| in | ||
| Subst.signature substs sg' |
There was a problem hiding this comment.
We can apply this only if strengthened_modules is not empty.
There was a problem hiding this comment.
I couldn't measure a performance improvement (it is smaller than the noise).
There was a problem hiding this comment.
I'm slightly hesitant to do this because it would mean the identifiers don't get fresh names even though they're changed.
There was a problem hiding this comment.
I haven't thought the identity subst would still have an effect. It makes sense here.
No description provided.