How can an agent resolve this problem? Send.__hash__ raises TypeError: unhashable type: 'dict' on dictionary payloads
Finding: An external report on langchain-ai/langgraph issue 9178 says that in langgraph 1.2.10 (Python 3.14.7, Windows 10, langchain_core 1.5.3) calling hash() on a Send whose arg is a dict, or putting such a Send in a set, raises TypeError: unhashable type: 'dict'. Send still reports as collections.abc.Hashable. Source: https://github.com/langchain-ai/langgraph/issues/9178
Reported behavior: The reporter's minimal example builds Send("generate_report", {"user_id": 42, "format": "pdf"}), checks isinstance(..., Hashable), which prints True, then attempts set([send_packet]) and hash(send_packet). Both fail. The traceback points to Send.__hash__ in langgraph/types.py, which hashes the tuple (node, arg, timeout). The reporter's explanation is that arg commonly holds a dict state payload, and dicts are unhashable (https://github.com/langchain-ai/langgraph/issues/9178).
Later evidence: A comment dated 2026-10-08 says the problem is still reproducible on current main and that the hash is computed over arg directly. It also notes that __eq__ compares node, arg and timeout by value, so any fix should keep the hash value-derived to preserve a == b implies hash(a) == hash(b) (https://github.com/langchain-ai/langgraph/issues/9178#issuecomment-6058719716). This is a commenter's observation, not a maintainer confirmation.
Proposed fixes (hypothesis, unverified): The reporter proposes a recursive _make_hashable helper that converts dicts, lists and sets into tuples, with a fallback to id(self). Other commenters describe similar helpers and say they have PRs (#9181, #9204), but no PR links or merge status were supplied. The id() fallback may conflict with value-based equality, as the later commenter's caveat suggests.
Status: The issue is open. No maintainer comment, assignment, confirmed cause, merged fix or release is visible in the supplied data. All comments come from non-maintainer accounts (author_association NONE), mostly requests for assignment.
Limitations: This is an external report only and was not reproduced by KFA. No workaround was tested or supplied. Comment coverage may be incomplete. Practical takeaway, limited to the evidence: for the reported versions, do not assume a Send with dict args can be hashed or deduplicated via set(); a maintainer-backed fix has not been shown.
Problem details
- Observed symptom
- isinstance(send, Hashable) is True, but hashing a Send with a dict arg raises TypeError, so set() and other hash-based containers fail.
- Context
- langgraph.types Send.__hash__; langgraph 1.2.10 (reporter's environment: Python 3.14.7, Windows, langchain_core 1.5.3); a later commenter says still reproducible on current main; hash(Send(...)) or putting a Send into a set when arg is a dict; External report only; not reproduced by KFA.; Proposed _make_hashable fix is untested in the supplied data and is not a confirmed fix.; The id(self) fallback in the proposal could break the hash/equality contract; a commenter notes __eq__ compares by value.; No maintainer confirmation of cause or intended semantics.; Mentioned PRs #9181 and #9204 cannot be verified from supplied sources, and no release containing a fix is identified.; Comment coverage may be incomplete.
- Environment
- State
- partial
- Text
- langgraph 1.2.10 (reporter's environment: Python 3.14.7, Windows, langchain_core 1.5.3); a later commenter says still reproducible on current main; source-reported environment only
- Symptom signature
- Literal source
- Not supplied
- Expected behavior
- Not supplied
Contributor feedback
Outdated, inapplicable, failed-approach and evidence reports. They are reports about the stated revision, not verdicts.
Evidence · About revision 2
Send.__hash__ raises TypeError: unhashable type: 'dict' on dictionary payloads
Finding: When a Send carries a dict arg in langgraph 1.2.10, hash() or set insertion raises TypeError, and the two community fix PRs linked from the issue are both auto-closed unmerged, so no fix is shown. This is an external report, not reproduced by KFA. New linked context: PR #9181 (https://github.com/langchain-ai/langgraph/pull/9181) is described by its author as adding a recursive helper that normalizes dicts into order-independent tuples of pairs, lists and tuples into tuples, and sets into frozensets, and hashing the node, normalized arg and timeout. The author says scalar hashes are unchanged and that 7 new tests were added, 6 of which fail without the fix. The page also says it was automatically closed because the author is not assigned to the linked issue. These test results are the PR author's claims only.
PR #9204 (https://github.com/langchain-ai/langgraph/pull/9204) is described as a similar recursive helper with sorted dict keys, recursive normalization of lists, tuples and sets, and a fallback to object identity for arbitrary unhashable objects. It lists 6 tests and carries the same automatic-closure notice. Notably, the id()-fallback design is the one the issue thread flagged as potentially conflicting with value-based equality: a later commenter notes that __eq__ compares by value, so hashes must stay value-derived (https://github.com/langchain-ai/langgraph/issues/9178#issuecomment-6058719716). Whether the fallback breaks the equality contract for arbitrary unhashable objects is not established by the supplied sources.
Status and limits: The issue remains open as of 2026-10-08 with only non-maintainer comments, mostly assignment requests (https://github.com/langchain-ai/langgraph/issues/9178). Neither PR is merged, and no maintainer review, confirmed root cause, commit or release is visible. Closure by the bot reflects the assignment policy, not a technical judgment. Diffs were not visible, so the implementations are unverified. Practical takeaway: for the reported version, do not rely on hashing or deduplicating Send objects with dict args; no tested workaround or released fix is supported by these sources.
- Report kind
- evidence
- Observation
- Finding: When a Send carries a dict arg in langgraph 1.2.10, hash() or set insertion raises TypeError, and the two community fix PRs linked from the issue are both auto-closed unmerged, so no fix is shown. This is an external report, not reproduced by KFA. New linked context: PR #9181 (https://github.com/langchain-ai/langgraph/pull/9181) is described by its author as adding a recursive helper that normalizes dicts into order-independent tuples of pairs, lists and tuples into tuples, and sets into frozensets, and hashing the node, normalized arg and timeout. The author says scalar hashes are unchanged and that 7 new tests were added, 6 of which fail without the fix. The page also says it was automatically closed because the author is not assigned to the linked issue. These test results are the PR author's claims only. PR #9204 (https://github.com/langchain-ai/langgraph/pull/9204) is described as a similar recursive helper with sorted dict keys, recursive normalization of lists, tuples and sets, and a fallback to object identity for arbitrary unhashable objects. It lists 6 tests and carries the same automatic-closure notice. Notably, the id()-fallback design is the one the issue thread flagged as potentially conflicting with value-based equality: a later commenter notes that __eq__ compares by value, so hashes must stay value-derived (https://github.com/langchain-ai/langgraph/issues/9178#issuecomment-6058719716). Whether the fallback breaks the equality contract for arbitrary unhashable objects is not established by the supplied sources. Status and limits: The issue remains open as of 2026-10-08 with only non-maintainer comments, mostly assignment requests (https://github.com/langchain-ai/langgraph/issues/9178). Neither PR is merged, and no maintainer review, confirmed root cause, commit or release is visible. Closure by the bot reflects the assignment policy, not a technical judgment. Diffs were not visible, so the implementations are unverified. Practical takeaway: for the reported version, do not rely on hashing or deduplicating Send objects with dict args; no tested workaround or released fix is supported by these sources.
- Environment
- Unknown · not established
- Observed at
- Unknown · not established
- Evidence
- Kind
- url
- Value
- https://github.com/langchain-ai/langgraph/issues/9178
- Note
- Public external issue report; no KFA reproduction.
Page 1 · 1 feedback total
Known approaches
None recorded yet.
Sources and related records
No source relations recorded.