Python: Prevent bad TC and add a bit of caching #5535
Merged
Conversation
This file contains hidden or 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
Using `simpleLocalFlowStep+` with the first argument specialised to `CfgNode` was causing the compiler to turn this into a very slowly converging manual TC computation. Instead, we use `simpleLocalFlowStep*` (which is fast) and then join that with a single step from any `CfgNode`. This should amount to the same thing. I also noticed that the charpred for `LocalSourceNode` was getting recomputed a lot, so this is now cached. (The recomputation was especially bad since it relied on `simpleLocalFlowStep+`, but anyway it's a good idea not to recompute this.)
A more principled approach is possible here, but in the short term this will prevent an explosion. For reference, openstack/cinder has roughly 19000 `ForTarget`s and tuples of size up to 5300, and we were calculating the cartesian product of these.
yoff
previously approved these changes
Mar 25, 2021
python/ql/src/semmle/python/dataflow/new/internal/DataFlowPrivate.qll
Outdated
Show resolved
Hide resolved
yoff
added
the
Awaiting evaluation
It's always a sad thing to see a good plan go wrong:
86860032 ~0% {4} r26 = JOIN r19 WITH DataFlowPublic::TupleElementContent#class#ff CARTESIAN PRODUCT OUTPUT Lhs.0 'nodeFrom', Lhs.1 'nodeTo', Rhs.0, Rhs.1
129256 ~3% {4} r27 = SELECT r26 ON In.3 <= 7
129256 ~0% {3} r28 = SCAN r27 OUTPUT In.0 'nodeFrom', In.2 'c', In.1 'nodeTo'
Happily, now it looks like this:
129256 ~0% {3} r20 = JOIN r19 WITH DataFlowPrivate::small_tuple#f CARTESIAN PRODUCT OUTPUT Lhs.0 'nodeFrom', Rhs.0, Lhs.1 'nodeTo'
yoff
previously approved these changes
Mar 25, 2021
RasmusWL
reviewed
Mar 26, 2021
| ) | ||
| } | ||
|
|
||
| pragma[noinline] | ||
| TupleElementContent small_tuple() { result.getIndex() <= 7 } |
There was a problem hiding this comment.
Did any special logic go into picking the value 7? or could it just as well have been 10? (either way, I would really appreciate a comment in the code explaining those details, so it's easy to figure out in 1 year time when we look at it again 😉 )
There was a problem hiding this comment.
I originally said 10, then yoff said that might be a bit high, then I said 5 but then changed my mind and said 7 which we both agreed was a nice round number.
I don't expect this code to be around in a year's time, and ideally not even a month from now. We already have a plan for a more principled approach for this (mimicking the "any index"-ness of list element content), but in the short term restricting this value is the best way to avoid blowup.
This one is a bit awkward, since the previous version was supposed to improve indexing. Unfortunately this is vastly outweighed by the slow convergence of the TC. Right now we pay the cost of inverting the `hasFlowSource` relation, but this is still cheaper.
yoff
approved these changes
Mar 27, 2021
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.
Using
simpleLocalFlowStep+with the first argument specialised toCfgNodewas causing the compiler to turn this into a very slowlyconverging manual TC computation.
Instead, we use
simpleLocalFlowStep*(which is fast) and then jointhat with a single step from any
CfgNode. This should amount to thesame thing.
I also noticed that the charpred for
LocalSourceNodewas gettingrecomputed a lot, so this is now cached. (The recomputation was
especially bad since it relied on
simpleLocalFlowStep+, but anywayit's a good idea not to recompute this.)