Globally optimize traversal in resolve #2454
Merged
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
Currently when we're attempting to resolve a dependency graph we locally optimize the order in which we visit candidates for a resolution (most constrained first). Once a version is activated, however, it will add a whole mess of new dependencies that need to be activated to the global list, currently appended at the end. This unfortunately can lead to pathological behavior. By always popping from the back and appending to the back of pending dependencies, super constrained dependencies in the front end up not getting visited for quite awhile. This in turn can cause Cargo to appear to hang for quite awhile as it's so aggressively backtracking. This commit switches the list of dependencies-to-activate from a `Vec` to a `BinaryHeap`. The heap is sorted by the number of candidates for each dependency, with the least candidates first. This ends up massively cutting down on resolution times in practice whenever `=` dependencies are encountered because they are resolved almost immediately instead of way near the end if they're at the wrong place in the graph. This alteration in traversal order ended up messing up the existing cycle detection, so that was just removed entirely from resolution and moved to its own dedicated pass. Closes rust-lang#2090
|
@bors r+ |
|
📌 Commit 4ec278f has been approved by |
bors
added a commit
that referenced
this pull request
Mar 12, 2016
Globally optimize traversal in resolve Currently when we're attempting to resolve a dependency graph we locally optimize the order in which we visit candidates for a resolution (most constrained first). Once a version is activated, however, it will add a whole mess of new dependencies that need to be activated to the global list, currently appended at the end. This unfortunately can lead to pathological behavior. By always popping from the back and appending to the back of pending dependencies, super constrained dependencies in the front end up not getting visited for quite awhile. This in turn can cause Cargo to appear to hang for quite awhile as it's so aggressively backtracking. This commit switches the list of dependencies-to-activate from a `Vec` to a `BinaryHeap`. The heap is sorted by the number of candidates for each dependency, with the least candidates first. This ends up massively cutting down on resolution times in practice whenever `=` dependencies are encountered because they are resolved almost immediately instead of way near the end if they're at the wrong place in the graph. This alteration in traversal order ended up messing up the existing cycle detection, so that was just removed entirely from resolution and moved to its own dedicated pass. Closes #2090
|
☀️ Test successful - cargo-linux-32, cargo-linux-64, cargo-mac-32, cargo-mac-64, cargo-win-gnu-32, cargo-win-gnu-64, cargo-win-msvc-32, cargo-win-msvc-64 |
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.
Currently when we're attempting to resolve a dependency graph we locally
optimize the order in which we visit candidates for a resolution (most
constrained first). Once a version is activated, however, it will add a whole
mess of new dependencies that need to be activated to the global list, currently
appended at the end.
This unfortunately can lead to pathological behavior. By always popping from the
back and appending to the back of pending dependencies, super constrained
dependencies in the front end up not getting visited for quite awhile. This in
turn can cause Cargo to appear to hang for quite awhile as it's so aggressively
backtracking.
This commit switches the list of dependencies-to-activate from a
Vecto aBinaryHeap. The heap is sorted by the number of candidates for eachdependency, with the least candidates first. This ends up massively cutting down
on resolution times in practice whenever
=dependencies are encounteredbecause they are resolved almost immediately instead of way near the end if
they're at the wrong place in the graph.
This alteration in traversal order ended up messing up the existing cycle
detection, so that was just removed entirely from resolution and moved to its
own dedicated pass.
Closes #2090