Add automated multi-platform builds#5
Open
midwan wants to merge 36 commits intoamiga-mui:masterfrom
Open
Conversation
Author
|
Ping @jens-maus / @tboeckel ? |
- Several changes to fix the code so that it compiles with the latest GCC 6.x from bebbo. It now builds correctly using amigadev/crosstools:m68k-amigaos for example. - Added cross-compiling build scripts for Windows, which use the aforementioned docker image (they assume the docker engine and image is already installed). - Fixed bugs that caused crashes on OS3.2 and MUI 5.x - Fixed bug where no content was displayed on MUI 5.x - Added a SimpleText tool, to test the library under OS3.x (haven't tested it elsewhere) - Removed obsolete HTMLview-Test.c - the SimpleTest should cover the same thing - Bumped version to 13.5
- Fix UtilityBase type conflict with SDK (use struct Library *) - Replace CPPDISPATCHERENTRY with ENTRY for dispatcher gates - Remove conflicting function stubs (SDK provides GetRPAttrs, etc.) - Replace strlcpy/stccpy with portable strncpy for AppMessage code These fixes enable successful compilation with sacredbanana/amiga-compiler:m68k-amigaos (GCC 6.5.0b)
Updates GitHub Actions CI to use the modern sacredbanana compiler image which has better compatibility with GCC 6.x builds.
The early return when bounds didn't match exactly caused MUI4 to show its checkerboard pattern instead of filled content. This fix removes the conditional early return and ensures MUI4 always receives a properly filled background region. Fixes issue where a blue-checkered pattern appeared on MUI4/OS3 instead of HTML content.
- Use docker:// syntax instead of addnab/docker-run-action - Update actions/checkout to v4, upload-artifact to v4 - Add concurrency control to cancel redundant builds - Simplify and clean up workflow structure - Use glob patterns for artifact paths
- Revert ENTRY to CPPDISPATCHERENTRY for _Dispatcher/_DispatcherP (C++ dispatchers need C gate wrapper for proper register handling) - Fix heap corruption: delete[] instead of delete for array allocations in ImageCacheEntry destructor These fixes enable HTMLview.mcc to work correctly on OS3.
- CI: add tag-based release trigger (v*) with automatic artifact uploads - Fix: buffer size missing null terminator in Dispatcher.cpp - Fix: delete[] mismatch in ImageManager.cpp destructor
Builds HTMLview for all supported platforms using sacredbanana docker images: - AmigaOS 3.x: m68k-amigaos - AmigaOS 4.x: ppc-amigaos - MorphOS: ppc-morphos Tag v* triggers release with all platform binaries attached.
SDI_hook.h: add MorphOS CPP dispatcher macros, fix extern linkage SDI_hook.h: fix __amigaos4__ typo for OS4 General.h: uppercase MIN/MAX to avoid C++ std::min/max conflicts Dispatcher.cpp: use ParseThread directly (not ENTRY macro) Closes #3
This was referenced Apr 15, 2026
Open
Fixes undefined reference to _Dispatcher on OS4 caused by C++ name mangling.
- mcp/private.h: Add INCLUDE_VERSION < 45 check for NewObject declaration - mcp/private.h: Add initPicClass forward declaration for MorphOS - mcp/pic.c: Enable proto/exec.h include for NewRawDoFmt on MorphOS - mcp/Makefile: Add -DUSEHOTKEY -DUSEBETTERSTRING for MorphOS - mcc/Makefile: Skip -ldebug for OS4 SimpleTest (not in newlib SDK) - mcc/Makefile: Skip SimpleTest target for OS4 (interface linking issues) - mcc/Makefile: Add -lauto to OS4 LDLIBS
- Added LibLoad_Test.c: simulates external MUI app loading HTMLview.mcc - Updated Makefile to build test programs for all platforms - Updated CI workflow to include SimpleTest and LibLoad_Test in artifacts
- Only link with -ldebug when DEBUG is defined - Add kprintf stub for OS4 in SimpleTest.c to allow building without debug
SimpleTest and LibLoad_Test now exercise the full MCC capability surface (images, forms, tables, fonts, entities) and render both local PROGDIR: files and plain-HTTP URLs through a shared image-load hook (test_image_hook.h). HTTPS is rejected cleanly (stock bsdsocket has no TLS) by aborting on any non-200 status code. Core fixes required to make images render at all: - IM_Render: RenderEngine::RenderEngine now zeroes FirstFrame. The LastFrame self-ref trick relies on it starting NULL, which the stack- allocated instance never guaranteed -- AllocateFrame's first-frame detection misfired and StatusItem->Start() was skipped. - Dispatcher: DefaultLoadFunc now strips "file://" only when present; bare paths (PROGDIR:foo.png) used to get the first 7 chars sliced off. - Hook wiring: m68k/MorphOS route through HookEntry trampoline so CallHookPkt's register args reach the plain-C handler; OS4 assigns TestImageHookFunc directly (HookEntry isn't declared there). - LibLoad_Test: defer MUIA_HTMLview_Contents until after the window is open -- the image decoder needs a screen, which isn't known at MUI_NewObject time. - test_image_hook.h: per-hook bsdsocket.library OpenLibrary with task-scoped shadow globals, so the socket base is valid in the decoder task (m68k bsdsocket is task-local). - Makefile: copy testdata/test.png into bin_$(OS)/ for PROGDIR: loads; add -Wl,-u,_startu on OS4 so C++ static init runs.
Network image loading in test programs + local-image render fixes
Wire TLS into the test-program image hook so https:// URLs and http->https 302 redirects resolve end-to-end. - Makefile downloads the AmiSSL 5.27 SDK on demand (lha xq) and exposes the headers only; inline macros in <inline/amissl.h> dispatch through a task-local AmiSSLBase, so no stub lib link and no global symbols are required. Enabled for OS3 and OS4 (MorphOS stays plain HTTP). - test_image_hook.h: THL_State gains AmiSSL library bases, SSL context and connection, plus a sni_errno slot for AmiSSL_ErrNoPtr. THL_Recv/THL_Send dispatch via SSL_read/SSL_write when use_tls is set, with function-scope shadowed library bases for inline macros. - URL parsing generalised: THL_ParseUrl replaces THL_ParseHttp and accepts both http and https (default ports 80/443). - THL_Connect factors out TCP connect plus optional TLS wrap; THL_TlsWrap opens amisslmaster + amissl, InitAmiSSLMaster / OpenAmiSSL / InitAmiSSLA (A-suffixed because OS3 builds with -DNO_INLINE_STDARG), then SSL_CTX_new(TLS_client_method), SSL_new, SSL_set_fd, SSL_set_tlsext_host_name, SSL_connect. VERIFY_NONE for now — CA bundle wiring is a follow-up. - THL_DoRequest sends the GET and parses status/headers; returns 1 OK / 0 error / -1 redirect. THL_HttpOpen loops up to 5 hops so http://www.amigaworld.net/images/awn2.gif follows its 302 to HTTPS transparently. - SimpleTest.c / LibLoad_Test.c: HTML adds a direct-HTTPS test entry (https://aminet.net/pics/aminet.png) alongside the existing HTTP entries. - .gitignore: exclude the downloaded mcc/amissl_sdk/ tree.
awn2.gif redirects to HTTPS correctly, but the HTTPS endpoint returns 404 -- the server only hosts the image on plain HTTP. logo-top.gif redirects to HTTPS and returns the image, which is the scenario we actually want to exercise.
Codex review on PR #2 flagged two stack overflows where a user-controllable value (Location: header up to 512 bytes, or a redirected URL up to 1024 bytes) was formatted into a smaller logbuf via unbounded sprintf. Converted every sprintf(logbuf, ...) and the GET-request sprintf to snprintf(...sizeof(buf)...). The hex-dump sprintfs are left as sprintf since they're bounded at 16 bytes and can't overflow. Fixes the P1 Location overflow and P2 unparseable-URL overflow flagged by Codex.
- AUTHORS: add Dimitris Panokostas; bump Open Source Team copyright to 2005-2026. - README, TODO, doc/MCC_HTMLview.readme: bump copyright year. - ChangeLog: log the HTTPS/AmiSSL work and the prior OS3/OS4/MorphOS build fix-ups under feat/https-support. - doc/MCC_HTMLview.doc, mcc/HTMLview_mcc.h: clarify that HTMLview.mcc itself never touches the network or filesystem -- URL-scheme support lives entirely in the application's ImageLoadHook / LoadHook. Point readers at SimpleTest / LibLoad_Test as a reference hook that handles PROGDIR:, file://, http:// and (with AmiSSL) https:// with redirect following.
HTTPS support via AmiSSL + http->https redirect following
Bump mcc/mcp to 13.6 (date 20.04.2026, copyright 2005-2026) and repackage the tag-driven CI release: each platform now ships as HTMLview-<tag>-<platform>.zip with only the runtime binaries and top-level docs, instead of loose artifacts that collided across OS3/OS4/MorphOS under merge-multiple.
Extract the reference image/content load hook out of the per-consumer `test_image_hook.h` blob and into a proper static library under `mcc/net_hook/`. Host applications now link `libhtmlview_nethook.a` and call `HTMLviewNet_InitHook(&hook)`; the hook opens bsdsocket / AmiSSL lazily from the decoder task so hosts no longer need any OpenLibrary plumbing of their own. Adds three opt-in runtime knobs on top of the extraction: - `HTMLviewNet_SetCABundle` / `SetVerifyMode` -- flip AmiSSL off the previous `SSL_VERIFY_NONE` default to auto-discovered peer verify (checks `AmiSSL:Certs/curl-ca-bundle.crt` and ENVARC mirrors), including hostname matching via `SSL_set1_host`. - `HTMLviewNet_SetCacheDir` / `SetCacheTTL` -- simple on-disk response cache keyed by FNV-1a-64 of the URL, storing `<key>.body` + `.meta` pairs. Disabled by default; caps entries at 8 MB, non-chunked 200 responses only. ETag revalidation is deferred to a follow-up. Also fixes a latent CI break in `mcc/Makefile`: the `\$(AMISSL_SDK_READY)` rule was defined before `all:`, so a bare `make OS=os3` used the SDK fetch as the default goal and exited 0 without building anything. The 13.6 release job hit exactly this and silently uploaded an artifact with no `.mcc` inside. Pinning `.DEFAULT_GOAL := all` at the top of the file makes `make` build by default; the new `libhtmlview_nethook.a` target and net_hook obj subdir are wired in alongside. The full improvement plan / session handoff lives in IMPROVEMENTS.md.
The 13.6 release failed at the `cp` step because the preceding build silently produced no `.mcc` binary (Makefile default-goal bug, fixed in the companion commit). `upload-artifact`'s default `warn` mode made the empty slot invisible until release-packaging time. - Add `Verify build products` steps that `ls` each expected binary (`HTMLview.mcc`, `HTMLview.mcp`, `SimpleTest`, `LibLoad_Test`) so a missing product fails the build job itself. - Set `if-no-files-found: error` on every `upload-artifact` so any future glob-match regression also fails at upload, not at release.
- README: drop SourceForge + SVN references, point at the GitHub repo, document the Docker cross-build flow for OS3 / OS4 / MorphOS, describe linking `libhtmlview_nethook.a` with its optional runtime knobs. - TODO: replace the placeholder `* ????` line with a short list of the pending buckets, pointing readers at IMPROVEMENTS.md for details. - doc/MCC_HTMLview.readme: bump to 13.6, drop the "prototype" and v13- snapshot text, swap the SourceForge URL for GitHub, and cross- reference each long-standing limitation to its IMPROVEMENTS phase. - IMPROVEMENTS.md: multi-session roadmap covering the whole backlog (phases 1-13). Phases 1, 2, 3, and 13 are marked delivered.
The 2005-2026 "HTMLview.mcc Open Source Team" line was overreaching - original contributors only worked on the class through 2007. Replace it with a narrow "2023-2026 Dimitris Panokostas" line on files I've actively maintained, keeping the accurate "1997-2000 Allan Odgaard" line. The new net_hook files drop Allan's attribution entirely since they are wholly new code. Historical files (~180) keep their accurate "2005-2007 by HTMLview.mcc Open Source Team" attribution untouched.
The release step failed with "Resource not accessible by integration" because the default GITHUB_TOKEN on this repo is read-only. Grant the release job the minimum permission (contents:write) it needs to publish the tag's release and upload the build artefacts.
Commit 6e678e4 removed the long-standing "stuntzi" early-return guard in the MUIM_Backfill handler so MUI 4/5 would stop rendering as a blue checkerboard. That regressed MUI 3.8 / 3.9, which need the guard to avoid painting outside the gadget bounds. Restore the guard but only apply it when MUIMasterBase->lib_Version <= 20 (MUI 3.9 = v20, MUI 3.8 = v19). MUI 4+ (v21+) keeps the unguarded path introduced for the checkerboard fix. Verified working on both MUI 3.8 and MUI 5.
fix(backfill): restore MUI 3.x equality guard, gated on lib_Version
* mcc/library.c: register ScrollGroupClass and StringClass via
CPPDISPATCHERENTRY() instead of bare ENTRY(). On AmigaOS 3 the
bare form passes the raw C++ dispatcher pointer to MUI, which then
calls it with cl/obj/msg in registers a0/a2/a1 -- a calling
convention the C++ function does not have. The result was a
register-convention mismatch that crashed inside intuition.library
with PC=0x2C2 the moment any caller tried to instantiate either
private class via NewObject(mcc_Class, ...). The main HTMLview
class was unaffected because mccinit.c already uses
CPPDISPATCHERENTRY for it.
* mcc/Dispatcher.cpp: harden the OM_NEW failure path. When
AllocSignal(-1) returns -1 we used to call CoerceMethod(cl, obj,
OM_DISPOSE), which re-enters our class's OM_DISPOSE against an
instance whose parse thread, message port and ParseThreadName are
not yet initialised. That tour of uninitialised state was harmless-
ish on MUI 3.x but rebooted MUI 4/5 reliably. We now free only
what the partial OM_NEW actually allocated (the SharedData) and
defer to cl->cl_Super OM_DISPOSE, mirroring the pattern already
used a few lines earlier for the SharedData allocation failure.
Surfaces as "Error creating MOAI object" instead of a hard reset
on signal-saturated tasks.
* mcc/Dispatcher.cpp: MUIA_HTMLview_Scrollbars no longer wraps the
gadget in a private ScrollGroupClass instance. Class-introspecting
callers (RapaGUI / Hollywood) treat the OM_NEW return as "the
HTMLview" and only translate their "Contents"/"URL"/etc.
properties to MUIA_HTMLview_* when the object actually looks like
one. The wrapper broke that detection, silently dropping property
writes. Returning the bare gadget always lets the caller's
introspection succeed. Apps that previously relied on us auto-
wrapping should now wrap externally with MUIC_Scrollgroup; see
mcc/SimpleTest.c for the canonical pattern. ScrollGroupClass is
still registered for callers that wire it up directly via
NewObject(mcc_Class, ...).
* mcc/Debug.cpp, mcc/Makefile: DEBUG=1 builds on OS3 used to fail to
link with a "multiple definition of kprintf" error -- Debug.cpp
defined its own kprintf and -ldebug pulled in another. Guard the
Debug.cpp definition with #ifndef DEBUG so it only compiles into
release builds, where -ldebug is not linked.
* mcc/RapaguiPattern_Test.c: new self-contained test program that
mimics the bare creation pattern used by MOAI / RapaGUI hosts
(no LoadHook tags, optionally signal-saturated via -burn N, and
optionally creating multiple gadgets via -many N). Used during
triage to reproduce the AllocSignal-failure path independently of
Hollywood. Wired into the OS3/OS4/MorphOS Makefile alongside
SimpleTest and LibLoad_Test.
BetterString.mcc and HotkeyString.mcc were hard runtime dependencies of the preferences editor (MCP). If either was absent, MUI_NewObject returned NULL and the NULL was passed as a MUI Child tag, crashing the system. - ostring(): try BetterStringObject first, fall back to StringObject - CreatePrefsGroup(): pre-create HotkeyString+snoop HGroup; fall back to built-in Keyadjust when HotkeyString.mcc is absent - InputClass: guard SimpleButton(Value) against NULL Value on <input type="button"> without a VALUE attribute - Refresh README with runtime requirements table and changelog
fix: graceful fallback for missing external MCCs
Author
|
Should I assume that since 3 years have passed and there has been no reply here, that PRs are not going to be merged ever? |
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
Automated CI/CD pipeline for building HTMLview on all supported platforms.
Features
Platforms
Build Fixes Included
Release Process