Repository navigation
Reinstall keeps drivers with the assign; wifipi uses the canonical extension header - #129
Merged
Merged
Conversation
fa671b0 made a drawer install write DEVICE=AmiNetXDuo:Devs/Networks/ for its drivers, but only for files it writes. A reinstall that answers "Keep it" leaves the old DEVICE=SYS:AmiNetXDuo/Devs/Networks/ anxnet.device in place (the lab A1200's file), and that drawer goes on opening the old drawer's drivers under a library from wherever the assign points next. The Installer cannot read a line back, so InstallSameFile gains a DEVICES mode: for each kept file in drawer layout it rewrites a DEVICE line only when it names anxnet/anxgenet/anxwifipi/anxzz9000.device by an absolute path whose directory is, by SameLock, this drawer's Devs/Networks or the one AmiNetXDuo: still names. Only the value changes; spacing, quotes, comments, other pairs and every other line stay byte-identical. A third-party driver, one of ours elsewhere, a relative name, a commented line or an escaped quoted value is left alone. The file is replaced through .new.info/.old.info as NetPrefs does, and a volume that is not mounted raises no requester. The rule is pure text in devicehome.c, covered on the host by test_devicehome (ctest devicehome_rewrites_only_ours). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
third_party/wifipi carries include/aminetxduo/anxs2ext.h, a hand-kept copy that src/wifipi/CMakeLists.txt put first. #126's callback pin reached the driver only when #128 patched the copy, and the copy still says ABI version 2: the driver tests Version == ANXD_S2_ABI_VERSION, the library sends 3 (9277398), so every open declined the extension and the single-copy receive never ran. The tree's include/ now comes first, so the driver reads the one header the library is built from. v2 to v3 changed only the ANXD_CMD_RX_BATCH record (Version/Size prefix), which the driver does not implement; the AnxdS2Extension prefix it reads is identical, and check-anxs2ext-abi now pins those offsets (0/2/4/8/12/16/20, size 24) for every version from MIN on. Codegen against the pinned toolchain: packet.c is instruction-identical, device.c differs only in the version constant (2 -> 3). Effect: anxwifipi.device accepts RX_DIRECT, RX_LINK_HDR and RX_VERIFIED from the library and the direct receive path runs for the first time since 2026-09-20, with no hardware run yet. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
WiFiPi maint/beta4 6594680 deletes its copy of anxs2ext.h, builds against ours, and accepts extension versions 2..3 as netdev_extension.c does, instead of exactly its copy's 2. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
replace_file() renamed a fresh file into place, so a kept interface file lost its protection bits and filenote. They are copied from the original onto the new file before the rename; a file that cannot be examined is replaced as before. devicehome.h named CheckNetConfig as a user; only InstallSameFile and its host test link it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The new host test registers on every host: 494 expected, 466 on darwin-arm64. Co-Authored-By: Claude Opus 5.5 <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.
Two beta7 defects found on the A1200 today.
1. A "Keep it" reinstall kept absolute driver paths (4d656aa). #127 fixed newly written interface files only. On a drawer-layout reinstall that keeps
Devs/NetInterfaces, aDEVICE=line naming one of our drivers by absolute path into this drawer (or the oneAmiNetXDuo:points at) kept opening that drawer's drivers. The Installer cannot read a file back, soInstallSameFile DEVICES(new mode of the existing helper) rewrites only such a line toAmiNetXDuo:Devs/Networks/<name>, by SameLock on the directory; everything else stays byte-identical (third-party drivers, our drivers elsewhere, relative and assign names, comments, quoting). Rule insrc/tools/devicehome.c, host testdevicehome_rewrites_only_ours(the lab A1200's exact file + 13 must-not-change lines). Not yet run on AmigaOS: needs codex's drawer-mode harness fix.2. anxwifipi.device compiled against a drifted copy of
anxs2ext.h(518404a, f00b33a). The fork's copy said extension version 2 and tested== 2; the library has offered 3 since 9277398 (2026-09-20, which changed only the RX_BATCH record, notAnxdS2Extension). So every wifipi since 09-20 declined the extension and ran the plain receive path. Now: the tree'sinclude/first for wifipi; the fork's copy deleted and the driver accepts versions 2..3 likenetdev_extension.c:11-12(WiFiPimaint/beta46594680);check-anxs2ext-abi.cmakepinsAnxdS2Extensionoffsets 0/2/4/8/12/16/20 and size 24. Layout identical under both headers;packet.ccodegen identical.Effect to validate on hardware: wifipi now accepts RX_DIRECT/RX_LINK_HDR/RX_VERIFIED, a path that has never run under
-mregparm. The A1200 Wi-Fi 49.5 Mbit/s figure was the fallback path; the final-archive hardware pass must re-measure it.tools/check-call-abi.sh: files=1831 findings=0.🤖 Generated with Claude Code