Skip to content

Python: Improve some flow summaries #22101

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
hvitved merged 2 commits into from
Jul 1, 2026
+34 −69
Merged

Python: Improve some flow summaries #22101

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a5444b5
Python: Improve some flow summaries
hvitved Jul 1, 2026
2bf6031
Python: Update inline test expectations
hvitved Jul 1, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion python/ql/lib/semmle/python/dataflow/new/internal/DataFlowPrivate.qll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1138,7 +1138,9 @@ predicate clearsContent(Node n, ContentSet cs) {
* Holds if the value that is being tracked is expected to be stored inside content `c`
* at node `n`.
*/
predicate expectsContent(Node n, ContentSet c) { none() }
predicate expectsContent(Node n, ContentSet c) {
FlowSummaryImpl::Private::Steps::summaryExpectsContent(n.(FlowSummaryNode).getSummaryNode(), c)
}

/**
* Holds if values stored inside attribute `c` are cleared at node `n`.
Expand Down
2 changes: 2 additions & 0 deletions python/ql/lib/semmle/python/dataflow/new/internal/FlowSummaryImpl.qll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -91,6 +91,8 @@ module Input implements InputSig<Location, DataFlowImplSpecific::PythonDataFlow>
cs.isAnyTupleOrDictionaryElement() and result = "AnyTupleOrDictionaryElement" and arg = ""
}

string encodeWithContent(ContentSet c, string arg) { result = "With" + encodeContent(c, arg) }

bindingset[token]
ParameterPosition decodeUnknownParameterPosition(AccessPath::AccessPathTokenBase token) {
// needed to support `Argument[x..y]` ranges
Expand Down
91 changes: 26 additions & 65 deletions python/ql/lib/semmle/python/frameworks/Stdlib.qll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4199,11 +4199,9 @@ module StdlibPrivate {
// The positional argument contains a mapping.
// TODO: these values can be overwritten by keyword arguments
// - dict mapping
exists(DataFlow::DictionaryElementContent dc, string key | key = dc.getKey() |
input = "Argument[0].DictionaryElement[" + key + "]" and
output = "ReturnValue.DictionaryElement[" + key + "]" and
preservesValue = true
)
input = "Argument[0].WithAnyDictionaryElement" and
output = "ReturnValue" and

@owen-mc owen-mc Jul 1, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This seems to change the meaning of the flow summary - previously it went from Argument[0].DictionaryElement[x] to ReturnValue.DictionaryElement[x] for all keys x, and now it goes from Argument[0].WithAnyDictionaryElement to ReturnValue. Can you explain?

@hvitved hvitved Jul 1, 2026

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Argument[0].DictionaryElement[x] is saying that data must be stored inside a dictionary value with some key, which has the exact same effect as the original flow summary, except it doesn't compile down to a bunch of read-steps followed by a bunch of identical store-steps, and it is hence much more performant.

@owen-mc owen-mc Jul 1, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, I think I get it now. I've never come across this mechanism before - go seems to be one of the few languages not using it. So, if I've understood correctly, if the input is <input>.WithAnyDictionaryElement and the output is <output> then the MaD machinery will create flow from <input> to <output> as long as there is flow to <input>.AnyDictionaryElement? A little bit like a lookahead in regexes, where it checks for the existence of something without consuming it.

preservesValue = true
or
// - list-of-pairs mapping
input = "Argument[0].ListElement.TupleElement[1]" and
Expand Down Expand Up @@ -4240,9 +4238,7 @@ module StdlibPrivate {
or
input = "Argument[0].SetElement"
or
exists(DataFlow::TupleElementContent tc, int i | i = tc.getIndex() |
input = "Argument[0].TupleElement[" + i.toString() + "]"
)
input = "Argument[0].AnyTupleElement"
// TODO: Once we have DictKeyContent, we need to transform that into ListElementContent
) and
// Element content is mutated into list element content
Expand All @@ -4266,11 +4262,9 @@ module StdlibPrivate {
}

override predicate propagatesFlow(string input, string output, boolean preservesValue) {
exists(DataFlow::TupleElementContent tc, int i | i = tc.getIndex() |
input = "Argument[0].TupleElement[" + i.toString() + "]" and
output = "ReturnValue.TupleElement[" + i.toString() + "]" and
preservesValue = true
)
input = "Argument[0].WithAnyTupleElement" and
output = "ReturnValue" and
preservesValue = true
or
input = "Argument[0].ListElement" and
output = "ReturnValue" and
Expand All @@ -4294,9 +4288,7 @@ module StdlibPrivate {
or
input = "Argument[0].SetElement"
or
exists(DataFlow::TupleElementContent tc, int i | i = tc.getIndex() |
input = "Argument[0].TupleElement[" + i.toString() + "]"
)
input = "Argument[0].AnyTupleElement"
// TODO: Once we have DictKeyContent, we need to transform that into ListElementContent
) and
output = "ReturnValue.SetElement" and
Expand Down Expand Up @@ -4342,9 +4334,7 @@ module StdlibPrivate {
or
input = "Argument[0].SetElement"
or
exists(DataFlow::TupleElementContent tc, int i | i = tc.getIndex() |
input = "Argument[0].TupleElement[" + i.toString() + "]"
)
input = "Argument[0].AnyTupleElement"
// TODO: Once we have DictKeyContent, we need to transform that into ListElementContent
) and
output = "ReturnValue.ListElement" and
Expand Down Expand Up @@ -4372,9 +4362,7 @@ module StdlibPrivate {
or
content = "SetElement"
or
exists(DataFlow::TupleElementContent tc, int i | i = tc.getIndex() |
content = "TupleElement[" + i.toString() + "]"
)
content = "AnyTupleElement"
|
// TODO: Once we have DictKeyContent, we need to transform that into ListElementContent
input = "Argument[0]." + content and
Expand Down Expand Up @@ -4404,9 +4392,7 @@ module StdlibPrivate {
or
input = "Argument[0].SetElement"
or
exists(DataFlow::TupleElementContent tc, int i | i = tc.getIndex() |
input = "Argument[0].TupleElement[" + i.toString() + "]"
)
input = "Argument[0].AnyTupleElement"
// TODO: Once we have DictKeyContent, we need to transform that into ListElementContent
) and
output = "ReturnValue.ListElement" and
Expand Down Expand Up @@ -4434,9 +4420,7 @@ module StdlibPrivate {
or
input = "Argument[0].SetElement"
or
exists(DataFlow::TupleElementContent tc, int i | i = tc.getIndex() |
input = "Argument[0].TupleElement[" + i.toString() + "]"
)
input = "Argument[0].AnyTupleElement"
// TODO: Once we have DictKeyContent, we need to transform that into ListElementContent
) and
output = "ReturnValue" and
Expand Down Expand Up @@ -4468,9 +4452,7 @@ module StdlibPrivate {
// We reduce generality slightly by not tracking tuple contents on list arguments beyond the first, for performance.
// TODO: Once we have TupleElementAny, this generality can be increased.
i = 0 and
exists(DataFlow::TupleElementContent tc, int j | j = tc.getIndex() |
input = "Argument[1].TupleElement[" + j.toString() + "]"
)
input = "Argument[1].AnyTupleElement"
// TODO: Once we have DictKeyContent, we need to transform that into ListElementContent
) and
output = "Argument[0].Parameter[" + i.toString() + "]" and
Expand Down Expand Up @@ -4499,9 +4481,7 @@ module StdlibPrivate {
or
input = "Argument[1].SetElement"
or
exists(DataFlow::TupleElementContent tc, int i | i = tc.getIndex() |
input = "Argument[1].TupleElement[" + i.toString() + "]"
)
input = "Argument[1].AnyTupleElement"
// TODO: Once we have DictKeyContent, we need to transform that into ListElementContent
) and
(output = "Argument[0].Parameter[0]" or output = "ReturnValue.ListElement") and
Expand All @@ -4525,9 +4505,7 @@ module StdlibPrivate {
or
input = "Argument[0].SetElement"
or
exists(DataFlow::TupleElementContent tc, int i | i = tc.getIndex() |
input = "Argument[0].TupleElement[" + i.toString() + "]"
)
input = "Argument[0].AnyTupleElement"
// TODO: Once we have DictKeyContent, we need to transform that into ListElementContent
) and
output = "ReturnValue.ListElement.TupleElement[1]" and
Expand All @@ -4552,12 +4530,7 @@ module StdlibPrivate {
or
input = "Argument[" + i.toString() + "].SetElement"
or
// We reduce generality slightly by not tracking tuple contents on arguments beyond the first two, for performance.
// TODO: Once we have TupleElementAny, this generality can be increased.
i in [0 .. 1] and
exists(DataFlow::TupleElementContent tc, int j | j = tc.getIndex() |
input = "Argument[" + i.toString() + "].TupleElement[" + j.toString() + "]"
)
input = "Argument[" + i.toString() + "].AnyTupleElement"
// TODO: Once we have DictKeyContent, we need to transform that into ListElementContent
) and
output = "ReturnValue.ListElement.TupleElement[" + i.toString() + "]" and
Expand All @@ -4580,12 +4553,6 @@ module StdlibPrivate {
override DataFlow::ArgumentNode getACallback() { none() }

override predicate propagatesFlow(string input, string output, boolean preservesValue) {
exists(DataFlow::Content c |
input = "Argument[self]." + c.getMaDRepresentation() and
output = "ReturnValue." + c.getMaDRepresentation() and
preservesValue = true
)
or
input = "Argument[self]" and
output = "ReturnValue" and
preservesValue = true
Expand Down Expand Up @@ -4741,12 +4708,10 @@ module StdlibPrivate {
override DataFlow::ArgumentNode getACallback() { none() }

override predicate propagatesFlow(string input, string output, boolean preservesValue) {
exists(DataFlow::DictionaryElementContent dc, string key | key = dc.getKey() |
input = "Argument[self].DictionaryElement[" + key + "]" and
output = "ReturnValue.TupleElement[1]" and
preservesValue = true
// TODO: put `key` into "ReturnValue.TupleElement[0]"
)
input = "Argument[self].AnyDictionaryElement" and
output = "ReturnValue.TupleElement[1]" and
preservesValue = true
// TODO: put `key` into "ReturnValue.TupleElement[0]"
}
}

Expand Down Expand Up @@ -4825,11 +4790,9 @@ module StdlibPrivate {
}

override predicate propagatesFlow(string input, string output, boolean preservesValue) {
exists(DataFlow::DictionaryElementContent dc, string key | key = dc.getKey() |
input = "Argument[self].DictionaryElement[" + key + "]" and
output = "ReturnValue.ListElement" and
preservesValue = true
)
input = "Argument[self].AnyDictionaryElement" and
output = "ReturnValue.ListElement" and
preservesValue = true
or
input = "Argument[self]" and
output = "ReturnValue" and
Expand Down Expand Up @@ -4876,11 +4839,9 @@ module StdlibPrivate {
}

override predicate propagatesFlow(string input, string output, boolean preservesValue) {
exists(DataFlow::DictionaryElementContent dc, string key | key = dc.getKey() |
input = "Argument[self].DictionaryElement[" + key + "]" and
output = "ReturnValue.ListElement.TupleElement[1]" and
preservesValue = true
)
input = "Argument[self].AnyDictionaryElement" and
output = "ReturnValue.ListElement.TupleElement[1]" and
preservesValue = true
or
// TODO: Add the keys to output list
input = "Argument[self]" and
Expand Down
4 changes: 2 additions & 2 deletions python/ql/test/library-tests/dataflow/coverage/test_builtins.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -589,11 +589,11 @@ def test_zip_tuple():

SINK(z[0][0]) # $ flow="SOURCE, l:-7 -> z[0][0]"
SINK(z[0][1]) # $ flow="SOURCE, l:-7 -> z[0][1]"
SINK_F(z[0][2])
SINK_F(z[0][2]) # $ SPURIOUS: flow="SOURCE, l:-7 -> z[0][2]"
SINK_F(z[0][3])
SINK(z[1][0]) # $ flow="SOURCE, l:-11 -> z[1][0]"
SINK_F(z[1][1]) # $ SPURIOUS: flow="SOURCE, l:-11 -> z[1][1]"
SINK(z[1][2]) # $ MISSING: flow="SOURCE, l:-11 -> z[1][2]" # Tuple contents are not tracked beyond the first two arguments for performance.
SINK(z[1][2]) # $ flow="SOURCE, l:-11 -> z[1][2]"
SINK_F(z[1][3])

@expects(4)
Expand Down
2 changes: 1 addition & 1 deletion python/ql/test/library-tests/frameworks/django-orm/testapp/orm_tests.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -362,7 +362,7 @@ def test_load_in_bulk():
# see https://docs.djangoproject.com/en/4.0/ref/models/querysets/#in-bulk
d = TestLoad.objects.in_bulk([1])
for val in d.values():
SINK(val.text) # $ MISSING: flow
SINK(val.text) # $ flow="SOURCE, l:-65 -> val.text"
SINK(d[1].text) # $ flow="SOURCE, l:-66 -> d[1].text"


Expand Down
Loading