Conversation
Bundles three things so there is one buildable state for regenerating cohort_fup (previously all of this was uncommitted working tree): - Empty-safe single-value FHIRPath functions, ported verbatim from 40190cc on fix/CQL-to-fhirpath-not-working-for-select (that branch does not contain the feature/segment tip, so the fix was unreachable from here). A bare X.round() raises in HAPI when the focus is empty, which is every calculatedExpression at render time - one of them makes the whole Questionnaire fail to render. Verified against HAPI org.hl7.fhir.r4 6.0.22: cohort_fup registration raised "focus for 0 can only have one value, but has 0 values" on bmi. See fix/20260831-fhirpath-empty-safe-math.md. - Choice/yes-no aware equality (fix/20260902-select-operand-coded-equality.md): a yes/no code literal compared with a native boolean item becomes a boolean literal, and an operand that only forwards a select's value (GetRepeatedValue, parenthesis) gets coded membership. An operand already reduced to a code (COALESCE union) keeps a plain '='. - The in-progress dynamic display text work (${REF} -> cqf-expression on item._text) that shares fhir_form.py, so this commit is self-contained. 409 tests pass.
`Length(a & b)` was emitted as `a & b.select($this.length())`. FHIRPath navigation binds tighter than `&`, so the call applied to `b` alone and the expression evaluated `string & integer`; HAPI raises "right operand to & has the wrong type integer", the exception leaves initializeCalculatedExpressions, and the whole Questionnaire fails to render. That is what crashed cohort_fup on the device (GlitchTip issue 97, "Failed to render questionnaire 50fb25b1-…"): `symptom_l` builds a comma-joined symptom label list guarded by `LENGTH(CONCATENATE(...)) > 0`, 18 times. Nothing raises while the form is empty, which is why it looked like "crashes when I select symptoms". Verified against HAPI org.hl7.fhir.r4 6.0.22 (the version OpenSRP 2.2.2-qa ships): the pushed registration Questionnaire raises with any answer present, and evaluates with 0 raises at 0/1/2/3 answers per repeating choice once the 18 sites are corrected. The mis-binding predates fix/20260831-fhirpath-empty-safe-math.md - a bare `.length()` mis-bound identically - so this is the same missing-parentheses bug, not a regression from it. See fix/20260902-fhirpath-suffix-precedence.md. 419 tests pass.
Large projects whose entry page chains many rhombus-guarded goto steps hung forever: expression generation re-expands each predecessor from scratch and expands every step's pnav_rel_* display bridge twice, so the chain costs ~2^n expansions of ever bigger expression trees. Python's recursion limit is never reached, so the run just never finished, with no log line naming a node. Add tricc_oo/visitors/loop_guard.py: LoopGuard (iteration budget for a while loop), RecursionGuard (depth + per-top-level-call budget, used as a context manager) and trip(), which logs the reason, the expansion path, the nodes revisited on it and a Python stack trace, then raises TriccLoopError. Guard targets are only turned into labels when a guard trips, so guarding a hot call stays cheap. Wire the guards into the two loops that could run away: get_node_expression is now a guarded wrapper around the unchanged body (_get_node_expression), and stashed_node_func caps its while loop -- check_stashed_loop only detects a frozen stash list, so an oscillating one could still spin forever. Budgets are tunable through TRICC_MAX_EXPRESSION_CALLS / _DEPTH, TRICC_MAX_LOOP_ITERATIONS and TRICC_LOOP_GUARD_PDB; the defaults leave three orders of magnitude over the reference projects' healthy peaks. Analysis in fix/20260902-expression-recursion-hang-guard.md, reading the failure in docs/troubleshooting.md. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A guard trip listed the whole expansion path and a revisit tally, which still left the reader to work out where the recursion actually turned back. Report it directly, narrowest first: the node the recursion looped on with the repeating segment around it, then the path from the nearest branching node down to that node -- the stretch of flowchart to open in draw.io. A repetition spanning other nodes (A -> B -> A) is a graph loop and wins over an immediate repeat (A -> A), which is instead reported as a re-entry that doubles the work per level -- what actually happens on a chain of rhombus-guarded goto steps. When no node repeats, the report says the recursion fans out and points at the deepest node. Branching means the graph forks at a node: more than one predecessor (the direction expansion walks) or more than one next node. The search starts strictly above the loop node so the segment always has a start; a loop node that branches itself is called out separately. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This branch has not been deployed
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Adding a guard to catch a looping and printing the trace where the loop occurred