jiic-iso build step fails on large staging trees (StringIndexOutOfBoundsException)

[LSantha]

Summary

The cd-x86-lite build fails inside the jiic-iso step (de.tu_darmstadt.informatik.rbg.hatlak.iso9660.ISOTask, wired in all/build-x86.xml:339-347) with java.lang.StringIndexOutOfBoundsException: String index out of range: 59 when the ISO staging tree contains on the order of ~1000 small files (a mauve testlet package tree staged under all/build/x86/cdrom-lite/ox/mauve/).

Environment

• JNode 0.2.9-dev tree, host JDK 1.6.0_45, sh build.sh cd-x86-lite
• Observed 2026-09-17

Reproduction

1. Populate the lite-CD staging dir with a deep tree of small files, e.g.:

   cp -r /tmp/jars/mauve/gnu/testlet/java/{io,lang,math,text,util} \
     all/build/x86/cdrom-lite/ox/mauve/gnu/testlet/java/
   # (~1000 .class files, ~4 MB total)
   

2. sh build.sh cd-x86-lite
3. Build fails:

   [jiic-iso] Rock Ridge support enabled.
   [jiic-iso] Joliet support enabled.
   [jiic-iso] El Torito support enabled.
   [jiic-iso] java.lang.StringIndexOutOfBoundsException: String index out of range: 59
       ... (ant stack through all/build-x86.xml:347 <- all/build-x86.xml:376 <- all/build.xml:591)
   BUILD FAILED
   

Removing the big tree (back to ~100 loose files) makes the same build succeed, so it is content/shape-dependent — likely the vendored ISO writer's 8.3 name-mangling or path handling tripping over a specific filename or the entry count (note the guest-side ISO9660 reader has a related deep-directory blind spot; the two may share filename assumptions).

Expected behavior

Either the tree is mastered correctly regardless of size/shape, or the task fails with a message naming the offending path (not a bare index exception).

Workaround

Keep staged file counts small and flat (we stage per-testlet loose files + one explicitly-listed sibling class instead of whole package trees).

Impact

Caps what can be shipped on the test CD (no bulk test corpora), forcing fragile per-file staging maintenance in local/mk-ox-iso.sh-style flows.

Triage Addendum (auto, 2026-09-24)

• Area / Kind: area/builder (best match) / kind/bug
• Blast radius: L5 system (ISO creation path; exact writer failure boundary unconfirmed)
• Expected scope: tentative: isolate and fix ISO writer path/name handling or improve failure reporting; exclude the guest ISO9660 reader pending reproduction
• Test path: sh build.sh cd-x86-lite with a deterministic large staging tree, then inspect/boot the generated ISO
• Merge signal: human-review (bug on the L5 ISO artifact path; writer source and failure boundary not yet established)
• Gaps: blocking: exact revision/clean setup, reproducible input source, full ISOTask stack, and host/backend selection
• Full report: ## Triage comment below.

---

✅ Ticket Runner Status

Field	Value
Phase	DONE
Turn	0/3
Retries	1/3
PR	-
Started	2026-09-25T05:57:21.178Z

[LSantha]

/oc triage

[LSantha]

Triage

• Area: area/builder (best match: the failure is inside builder/lib/iso9660.jar ISOTask, invoked by all/build-x86.xml:339-347)
• Kind: kind/bug
• Repro: needs more info from reporter
• Blast radius: L5 system (ISO creation path; an invalid or unmasterable boot ISO can block distribution artifacts, while the suspected failure mechanism remains unconfirmed)
• Expected scope: tentative: isolate and fix ISO writer path/name handling, or improve failure reporting at the exact failing method; no guest ISO9660 reader or unrelated build targets pending reproduction
• Test path: sh build.sh cd-x86-lite with the reported large staging tree, followed by inspection/boot of the generated ISO once the deterministic input is available
• Merge signal: human-review (bug on the L5 ISO artifact path; exact writer source and failure boundary are not yet established)
• Gaps: blocking: exact revision and clean setup; reproducible input source; full writer stack; host/backend selection. Non-blocking: no split indicated
• Suggested next: needs-info
• Split proposal: none yet; split only if the eventual writer fix and guest reader issue prove independent and both are brought into scope
• Labels applied: kind/bug, area/builder, no-auto

Needs the following before work can start:

1. @reporter: provide git log -1 --oneline, host OS, and whether the tree was clean before staging; confirm the exact sh build.sh cd-x86-lite command and host JDK.
2. @reporter: provide a deterministic input that another maintainer can obtain from the repository or a stable artifact (for example, an archive or exact file-list generator) instead of /tmp/jars/mauve.
3. @reporter: include the full Ant/ISOTask stack excerpt (about 40-60 lines), including the first vendored de.tu_darmstadt... frame; identify the offending path or filename if Ant exposes it.
4. @reporter: confirm why the jiic-iso backend was selected (Windows or unavailable mkisofs) and whether the same tree succeeds through the mkisofs backend.

Auto-run: skipped (no-auto)

[LSantha]

#648 (comment)

opencode session  |  github run

[LSantha]

/run

[LSantha]

/oc Please proceed with this task.

[LSantha]

Working on this. Run: 36100738078
Plan: recover the vendored ISOTask source, reproduce the large-tree boundary, patch and regression-test name/path handling, then run the lite ISO build.

[LSantha]

/oc Please proceed with this task.

[LSantha]

Created PR #668

opencode session  |  github run