vms-21a: STARTUP.COM becomes a real STDRV phased startup driver - #286
Merged
Merged
Conversation
Core mechanism, oracle-pinned to the Alpha V8.4 boot capture (docs/design-boot-faithful.md sec3.2/3.3): - A real [SYS0.SYSCOMMON.SYS$STARTUP] directory (ovmx_layout.h VMS_SYS_STARTUP) carries SYS$STARTUP:VMS$PHASES.DAT (the nine OBSERVED phase names verbatim: INITIAL DEVICES PRECONFIG CONFIG BASEENVIRON LPBEGIN LPMAIN LPBETA END) and SYS$STARTUP:VMS$VMS.DAT (an OVMX-DEFINED component-registration format, Rule 8 -- only the VSI file NAME is observed, the line format inside is OVMX's own invention). - SYS$STARTUP is now the measured multi-value SEARCH LIST logical (DEFINE/SYSTEM SYS$STARTUP dir,SYS$MANAGER -- exactly what vms-420 built DEFINE/SYSTEM multi-value support for). OVMX flattens the first element the same way lnm_defaults.c already flattens SYS$MANAGER (straight off SYS$SYSDEVICE rather than via a rooted SYS$SYSROOT), and STARTUP.COM's own comment flags that divergence from the oracle's literal "SYS$SYSROOT:[SYS$STARTUP]" text. - STARTUP.COM is now a GOSUB-based phase driver (RUN_PHASES): reads VMS$PHASES.DAT, and for each phase runs (a) every component VMS$VMS.DAT registers for it (RUN_COMPONENTS -- none registered yet, so every phase currently runs empty, which is honest per the design doc's "empty phases run empty" target) and (b) the one site file this project's reading of the OpenVMS System Manager's Manual assigns to that phase: SYCONFIG.COM at CONFIG, SYLOGICALS.COM at BASEENVIRON, SYSTARTUP_VMS.COM (unchanged content, only its call site moved) at LPMAIN. Verified end-to-end against a local build of DCL.EXE (process- scope logicals, since SYSTEM-scope needs the real executive) before wiring into the shipped file -- OPEN/READ/GOSUB/F$ELEMENT/EOF-via- $STATUS all confirmed correct. - SYS$MANAGER:OVMX.CONF and SYLOGICALS.CONF deleted: both verified to have ZERO code readers on main (OVMX.CONF was never referenced by any constant; SYLOGICALS.CONF's only reference, VMS_LNM_CONF_PATH, was itself dead and is removed). A real VMS site file is a DCL command procedure, not a KEY=VALUE config file, per design-boot-faithful.md sec2.4. NODE_NAME already maps to SCSNODE (vms-b6a7, unchanged by this commit); VERSION_LIMIT/DEFAULT_PROTECTION had no live OVMX SYSGEN parameter or SET command to map to, so they are dropped rather than ported to a home that doesn't exist yet. - SYCONFIG.COM and SYLOGICALS.COM are new, minimal, OVMX-authored site files (tools/ovmx_kit_pack.c already listed both as seed-once filenames, anticipating this work). ovmx_banner.h and test_identity_ssot.sh updated to point at SYLOGICALS.COM instead of the deleted .CONF. Dockerfile.bootable/CI wiring for the new directory and a QEMU boot proof follow in a subsequent commit on this branch. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ver boot proof Dockerfile.bootable: the FAT initramfs, kit-stage, and mastered distrib image all now carry [SYS0.SYSCOMMON.SYS$STARTUP] (VMS$PHASES.DAT, VMS$VMS.DAT) alongside the existing SYSEXE/SYSLIB/SYSMGR/SYSHLP/SYSUPD directories -- without this copy STARTUP.COM's new phase-driver OPENs fail %RMS-E-FNF on every boot. Added build-time gates on the mastered distrib image: VMS$PHASES.DAT/VMS$VMS.DAT/SYCONFIG.COM/SYLOGICALS.COM must be present, and OVMX.CONF/SYLOGICALS.CONF must NOT be -- a regression that recreates either deleted file now fails the build, not just a runtime check. tests/qemu/test_startup_phase_driver.sh (new): boots the pre-installed distribution disk, logs in over the console, and proves against the REAL running system (not a build-time inspection) that SHOW LOGICAL SYS$STARTUP reports the measured multi-value search list (SYS$MANAGER as its second element -- vms-420's multi-value DEFINE/SYSTEM actually ran at boot), SYS$STARTUP:/SYS$MANAGER: carry the new data/site files, and DIRECTORY SYS$MANAGER:OVMX.CONF finds nothing on the booted, mounted disk. Also checks the pre-login boot console for the LPMAIN-phase site announcement and the absence of %RMS-E-FNF/ %DCL-E-OPENIN during boot (scoped to the boot-only log segment, so the later intentional OVMX.CONF absence-check can't be mistaken for a regression). Registered in ci.yml's persistent-boot job, right after the existing distrib-boot test it reuses the built image from. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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.
Summary
STARTUP.COMis now STDRV, a phased startup driver, not a flat script. It reads the nine OBSERVED phase names (verbatim from the Alpha V8.4 oracle capture) from a realSYS$STARTUP:VMS$PHASES.DAT, and for each phase runs the OVMX-defined components registered inSYS$STARTUP:VMS$VMS.DAT(none registered yet — every phase currently runs empty, which is honest) plus the one documented OpenVMS site file this project's reading of the System Manager's Manual assigns to it:SYCONFIG.COMat CONFIG,SYLOGICALS.COMat BASEENVIRON,SYSTARTUP_VMS.COM(unchanged content, only its call site moved) at LPMAIN.SYS$STARTUPis now the measured multi-value search-list logical (DEFINE/SYSTEM SYS$STARTUP dir,SYS$MANAGER) — exactly what vms-420 built DEFINE/SYSTEM multi-value support for. A real[SYS0.SYSCOMMON.SYS$STARTUP]directory backs it (ovmx_layout.hVMS_SYS_STARTUP). OVMX flattens the first element the same waylnm_defaults.calready flattensSYS$MANAGER(offSYS$SYSDEVICEdirectly, not via a rootedSYS$SYSROOT) — flagged explicitly inSTARTUP.COM's own comments as a divergence from the oracle's literalSYS$SYSROOT:[SYS$STARTUP]text.SYS$MANAGER:OVMX.CONFandSYLOGICALS.CONFare deleted — both verified to have zero code readers on main. A real VMS site file is a DCL command procedure, never aKEY=VALUEconfig file.NODE_NAMEalready maps toSCSNODE(vms-b6a7, unchanged here);VERSION_LIMIT/DEFAULT_PROTECTIONhad no live OVMX SYSGEN parameter or SET command to map to, so they're dropped rather than ported to a home that doesn't exist yet.SYCONFIG.COMandSYLOGICALS.COM(the kit packer already listed both as seed-once filenames, anticipating this work).Dockerfile.bootablewired to ship the new[SYS$STARTUP]directory in the FAT initramfs, the OS kit, and the mastered distrib image, with build-time gates asserting the new files are present and the deleted.CONFfiles are absent from the mastered image.tests/qemu/test_startup_phase_driver.sh, wired into CI'spersistent-bootjob.Clean-room (Rule 8): phase names / file names are OBSERVED (Alpha V8.4 oracle capture,
docs/design-boot-faithful.md§3.2/§3.3);VMS$VMS.DAT's line format is an OVMX invention, labeled as such; no VSI.COM/.EXEcontent was read.Ground-source proof (real QEMU, real vms.ko + vmsfs.ko)
Ran all three locally against a from-scratch
distro/Dockerfile.bootablebuild (hardtimeout-wrapped, no host installs):tests/qemu/test_startup_phase_driver.sh— 19/19 passed. Boots the mastered distrib image, logs in, and proves against the real running system:SHOW LOGICAL SYS$STARTUPreports the multi-value search list withSYS$MANAGERas its second element;SYS$STARTUP:/SYS$MANAGER:carry the new data/site files;DIRECTORY SYS$MANAGER:OVMX.CONFfinds nothing; the LPMAIN-phase site announcement appears; no%RMS-E-FNF/%DCL-E-OPENINduring boot.tests/qemu/test_persistent_boot.sh(mount-or-halt, PRODUCT INSTALL: land a real kit on a real vmsfs volume (vms-df9) #278's norecord-halt regression) — 20/20 passed, no regression.tests/qemu/test_distrib_boot.sh— 17/17 passed, no regression.The DCL phase-loop logic itself (
OPEN/READ/GOSUB/F$ELEMENT/EOF-via-$STATUS) was also verified standalone against a local build ofDCL.EXEbefore being wired into the shipped file.Test plan
tests/qemu/test_startup_phase_driver.sh(new) — 19/19 in QEMUtests/qemu/test_persistent_boot.sh— 20/20, no regressiontests/qemu/test_distrib_boot.sh— 17/17, no regressiontests/integration/test_identity_ssot.shupdated to checkSYLOGICALS.COMinstead of the deleted.CONFCo-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com