Branches performance
Pull requests
last run
14 hours ago
Fix printer for functions.
We would previously inject a redundant `function` before dispatching to
the printer of implementations which already emits the kind.
14 hours ago
47937a1
topic/bbannier/function-print
0%
last run
3 days ago
Add env var to skip running `FileCheck` tests
Since `FileCheck` may give verbose output, and is not updated with
baselines (ie with -U), it may be useful during development to skip the
`FileCheck` tests, then later run them when it is closer to completion.
3 days ago
9b796f2
topic/etyp/filecheck-tests-hardening
0%
last run
3 days ago
Optimize grouping with side-effect free local variables.
Our resolver can sometimes leave new groupings behind that are more
complex than necessary because they introduce local temporary
variables even though their value is constant and side-effect free.
The problem is that the resolver can't tell if something is
side-effect free (that needs data flow analysis) and hence it needs to
remain conservative.
This commit adds a peephole optimization that removes local variables
from groupings if they aren't necessary. We generally prefer repeating
the side-effect expressions for readability and further optimizations
down the line.
3 days ago
40585b4
topic/robin/opt-groupings
0%
last run
3 days ago
Make the visitors' `replaceNode()` safe.
The existing `replaceNode()` method was impossible to use safely: if
it was passed a replacement node that was already part of the AST
somewhere, the method would disconnect that node from its original
position first. That, however, is generally not safe because AST nodes
do not give any guarantees about their internal child layout, so that
simply removing one can lead to trouble. Plus, one always had to keep
mind that the replacement node would now disappear from its original
place, which is error-prone even in safe cases.
This changes `replaceNode()` to instead use our standard semantics
when making AST modifications: If a node is being inserted that
already has a parent, we deep-copy it first. That way, the caller
doesn't need to worry about safe memory management. In addition, we
add a new method `replaceNodeWithChild()` that optimizes the operation
for a special case: If an existing child is taking the position of a
parent of itself, then we can always safely move it over into its new
place without copying. We now use that for cases across the code base
that match this special case.
5 days ago
12e4ee6
topic/robin/replace-node
0%