Skip to content

Add release pipeline: working installer, CI, and distribution docs - #1

Merged
ShixSun merged 2 commits into
mainfrom
release-infrastructure
Aug 6, 2026
Merged

ShixSun merged 2 commits into
mainfrom
release-infrastructure

Conversation

@ShixSun

@ShixSun ShixSun commented Aug 6, 2026

Copy link
Copy Markdown
Owner

Amber had no way for anyone to actually install it — no Releases, no installer, and the packaging work that existed was uncommitted and unbuildable. This makes the project distributable end to end.

Installer (installer/Amber.nsi) — was a scaffold whose detection logic was placeholder code. DetectFoobar2000Page hardcoded C:\Program Files\ foobar2000 (with a "REMOVE THIS before production" note) and ComponentCheckPage unconditionally set JSPLITTER_OK=1, so a compiled build would have installed to a fixed path regardless of where foobar2000 actually lives. Implemented for real:

  • locates foobar2000 via registry, App Paths, then conventional dirs, with a Browse fallback and live validation feedback
  • rejects pre-2.0 via the exe's version resource
  • resolves portable vs roaming profiles, including v1-era profile dirs
  • probes user-components-x64/foo_uie_jsplitter, which settles both architecture and the required-component check in one test; an x86 profile now gets a specific message instead of "component missing"
  • reports optional components without blocking
  • backs up theme.fth and any previous Amber scripts before writing
  • checks profile writability up front rather than failing mid-install
  • uninstaller parses its recorded metadata instead of guessing the profile from $INSTDIR..

theme.fth is still never overwritten — importing the layout is one click in foobar2000, while a bad automatic overwrite costs the user a layout they may have spent hours on.

Build fixes:

  • OutFile pointed at ....\dist, one level above the File commands' ..\dist, so artefacts landed outside the repo entirely
  • branding assets were never generated, so MUI_ICON broke the compile
  • the version was hardcoded in the NSI, contradicting manifest.json's own single-source-of-truth rule; it now comes from a generated dist/version.nsh, and validate-release.py fails the build if a literal reappears
  • SHA256SUMS was written before makensis ran, so it never covered the installer; --checksums-only recomputes it afterwards

CI: ci.yml validates and stages on Linux and compiles the installer on Windows, so packaging breakage surfaces on push rather than at tag time. release.yml publishes on tag, refusing to build if the tag and manifest.json disagree.

Docs: README leads with a download section, INSTALL.md documents the installer path, and THIRD_PARTY_LICENSES.md records why no component is bundled — freely downloadable is not freely redistributable.

Verified locally: validation passes 6/6, staging produces a clean 13-entry theme zip with no build artefacts, and makensis compiles a 544 KB installer. The detection logic has not yet been exercised against a real foobar2000 install on Windows.

ShixSun and others added 2 commits August 6, 2026 11:10
Amber had no way for anyone to actually install it — no Releases, no
installer, and the packaging work that existed was uncommitted and
unbuildable. This makes the project distributable end to end.

Installer (installer/Amber.nsi) — was a scaffold whose detection logic
was placeholder code. DetectFoobar2000Page hardcoded C:\Program Files\
foobar2000 (with a "REMOVE THIS before production" note) and
ComponentCheckPage unconditionally set JSPLITTER_OK=1, so a compiled
build would have installed to a fixed path regardless of where
foobar2000 actually lives. Implemented for real:

  - locates foobar2000 via registry, App Paths, then conventional dirs,
    with a Browse fallback and live validation feedback
  - rejects pre-2.0 via the exe's version resource
  - resolves portable vs roaming profiles, including v1-era profile dirs
  - probes user-components-x64/foo_uie_jsplitter, which settles both
    architecture and the required-component check in one test; an x86
    profile now gets a specific message instead of "component missing"
  - reports optional components without blocking
  - backs up theme.fth and any previous Amber scripts before writing
  - checks profile writability up front rather than failing mid-install
  - uninstaller parses its recorded metadata instead of guessing the
    profile from $INSTDIR\..

theme.fth is still never overwritten — importing the layout is one click
in foobar2000, while a bad automatic overwrite costs the user a layout
they may have spent hours on.

Build fixes:
  - OutFile pointed at ..\..\dist, one level above the File commands'
    ..\dist, so artefacts landed outside the repo entirely
  - branding assets were never generated, so MUI_ICON broke the compile
  - the version was hardcoded in the NSI, contradicting manifest.json's
    own single-source-of-truth rule; it now comes from a generated
    dist/version.nsh, and validate-release.py fails the build if a
    literal reappears
  - SHA256SUMS was written before makensis ran, so it never covered the
    installer; --checksums-only recomputes it afterwards

CI: ci.yml validates and stages on Linux and compiles the installer on
Windows, so packaging breakage surfaces on push rather than at tag time.
release.yml publishes on tag, refusing to build if the tag and
manifest.json disagree.

Docs: README leads with a download section, INSTALL.md documents the
installer path, and THIRD_PARTY_LICENSES.md records why no component is
bundled — freely downloadable is not freely redistributable.

Verified locally: validation passes 6/6, staging produces a clean
13-entry theme zip with no build artefacts, and makensis compiles a
544 KB installer. The detection logic has not yet been exercised against
a real foobar2000 install on Windows.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
GitHub Windows runner does not have makensis in PATH by default.
Install it via Chocolatey before the compile step.
@ShixSun
ShixSun merged commit 2d94d11 into main Aug 6, 2026
1 of 2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant