RFC: Also change lowering of getproperty #37289
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
In expressions like `Foo.Bar.f() = 1`, we would lower this as `Expr(:method, Expr(:call, :getproperty, Expr(:call, :getproperty, :Foo, :Bar), :f) ...)` where args[1] is supposed to be the name of the method. However, if this is not a symbol, the name is actually ignored and the method is required to exist already (i.e. `function foo(); end` is allowed to introduce a new method if `foo` doesn't exist, but `function Main.foo(); end` is not). It doesn't do much harm currently other than being a bit ugly. However, I was trying to do some experiments with getproperty lowering that made the lowering a bit more complicated, which made the frontend unhappy, because there were now more complicated expressions in there (the fact that it works for call is a bit of an accident). Since there is no good reason to keep it and I did the work to drop it anyway, for my experimenting, let's just drop this unused expr arg entirely and replace it by nothing if it's not supposed to be used.
So after thinking about #37268, I was wondering what would happen if one applied the same trick to the lowering of `getproperty` as well. This PR starts at that, but is mostly a placeholder for discussion of whether this is something we want to do. In particular, this PR lowers `a.b` to: ``` getproperty(a)(a, :b) ``` with a default implementation of ``` getproperty(a) = getfield ``` The inference benefit is as expected. Timing end-to-end inference&optimize time of: ``` struct Foo; end f(a::Foo) = a.x ``` shows that time to infer `f` drops by about a third (because it doesn't need to infer getproperty twice). Overall build time of the system image drops by 6% for me. If we do decide to go this way, things that would still need to be fixed: - [] Do the same for `setproperty!` - [] `ccall` doesn't like this lowering. Didn't look into it too closely - [] Needs a fallback to backwards compatibility. I'm thinking something along the lines of ``` getproperty(x) = hasmethod(getproperty, Tuple{typeof(x), Symbol}) ? getproperty : getfield ``` but obviously it would need inference integration to be fast.
|
In discussion with @JeffBezanson and @vtjnash we decided this was at least a worthwhile direction to explore, so we'll get this PR in shape. @JeffBezanson wanted to look at the ccall lowering issue and then I can finish this up. |
|
We had a similar discussion for |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
So after thinking about #37268, I was wondering what would happen
if one applied the same trick to the lowering of
getpropertyas well. This PR starts at that, but is mostly a placeholder for
discussion of whether this is something we want to do. In particular,
this PR lowers
a.bto:with a default implementation of
The inference benefit is as expected. Timing end-to-end inference&optimize
time of:
shows that time to infer
fdrops by about a third (because it doesn't need toinfer getproperty twice). Overall build time of the system image drops
by 6% for me.
If we do decide to go this way, things that would still need to be fixed:
setproperty!ccalldoesn't like this lowering. Didn't look into it too closelyalong the lines of
but obviously it would need inference integration to be fast.