You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
cobol-to-java: same-name programs collapse into one Service/Controller (the Java side of #3218) #3221
#3218 (fixed in #3219) keyed the clean-room outputs per path. When two programs share a stem, their IR, JCL and schema files become COBOL__SAM2_* and multiroot__sam__SAM2_*.
gitgalaxy/cobol_to_java_controller.py still derives each Java class name from the IR's metadata.file_name: SAM2.cbl becomes Sam2Service.java and Sam2Controller.java. Both SAM2 programs therefore write the same Java files, and the later one in IR-filename order overwrites the earlier.
Evidence. IBM/zopeneditor-sample @ 8f983530:
stage
what it holds
clean room
5 IR dumps: COBOL__SAM1, COBOL__SAM2, multiroot__sam__SAM1, multiroot__sam__SAM2, SAM1LIB
The #3212 PR commits the snapshot: tests/cobol_mainframe/refraction_snapshot/excerpts/zopeneditor-sample/java/.
Entities can collide the same way, because they are named from the schema title.
Direction. Name each Java class from the IR dump's file key (the stem of <key>_ir.json), not from metadata.file_name. Bless the refraction snapshot in the same PR (python tests/tools/refraction_snapshot.py update, plus --corpus zopeneditor-sample locally).
Fixed in PR #3231. Scope note: the issue's evidence was 5 IR dumps → 3 services; the entity half turned out to be much larger. DFHCOMMAREA is every CICS program's record title, so CBSA was generating 3 entities for 29 schemas and carddemo 13 for 36 — 54 programs' data layouts discarded across the three corpora, now 1:1. Uncovered #3233 (26 entity classes now each declare @Table(name = "DFHCOMMAREA")) and an agent-job key bug (slice_file.name.split("_")[0] returned COBOL for COBOL__SAM2_slice.json), fixed in the same PR.
Activity
fix(refraction): case-sensitive sort keys (Windows), non-overlapping …
test(refraction): pinned mainframe corpora, generated-output snapshot…
squid-protocol commentedon Sep 19, 2026
Fixed in PR #3231. Scope note: the issue's evidence was 5 IR dumps → 3 services; the entity half turned out to be much larger.
DFHCOMMAREAis every CICS program's record title, so CBSA was generating 3 entities for 29 schemas and carddemo 13 for 36 — 54 programs' data layouts discarded across the three corpora, now 1:1. Uncovered #3233 (26 entity classes now each declare@Table(name = "DFHCOMMAREA")) and an agent-job key bug (slice_file.name.split("_")[0]returnedCOBOLforCOBOL__SAM2_slice.json), fixed in the same PR.fix(refraction): name every generated Java class from the clean-room …
fix(refraction): a DFHCOMMAREA maps to a DTO, not a JPA entity (#3233) (