* fix(start-scripts): find bundled privacy_core.dll next to script
start.bat and start.sh only checked the source-tree DLL path
(``privacy-core/target/release/privacy_core.dll``), not the bundled
location where MSI/AppImage/DMG installers stage the library directly
next to the script in backend-runtime/.
Users running start.bat from inside an MSI install dir (a documented
workaround when the desktop shell crashes) saw a scary "install Rust"
warning even though the DLL was sitting right next to them. See issue
#319 for the user-reported confusion.
Fix: add a fallback check for the bundled location before falling
through to the "build privacy-core from source" warning. Source-tree
behavior unchanged — the source path is still preferred when present.
Also re-stamps the v0.9.81 source archive: ``release_digests.json``
v0.9.81 zip hash updated to point at the rebuilt source archive that
contains these script changes. MSI/EXE/sig hashes are unchanged (the
scripts live at the repo root, not inside the desktop bundle).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(#319): bundle start.bat + start.sh into the MSI/EXE installers
Follow-up to the start-script DLL fallback fix in the prior commit.
ChrisMTheMan's report on #319 made it clear the workaround flow was:
1. MSI install crashes on launch (different bug, fixed in v0.9.81)
2. User goes looking for start.bat to launch the backend manually
3. start.bat isn't in their install dir, so they go fetch it from GitHub
4. They get a working script but it doesn't know about the bundled
privacy_core.dll layout, so they see a scary "install Rust" warning
The prior commit fixed step 4. This commit fixes step 3 — start.bat and
start.sh now ship inside the MSI/EXE installers (staged into
backend-runtime/ next to the privacy_core.dll they expect to find).
After the rebuild lands, an MSI user looking for these scripts finds
them right inside their install dir, already pointing at the correct
bundled DLL location.
What changed
------------
* ``build-backend-runtime.cjs`` now has a ``stageStartScripts()`` step
that copies start.bat and start.sh from the repo root into the
staged backend-runtime/. Preserves the executable bit on .sh under
POSIX.
* ``release_digests.json`` v0.9.81 block hashes refreshed for the
rebuilt MSI / EXE / source-zip (the scripts being bundled changed
the MSI/EXE contents; the source zip also includes the start-script
fix from the prior commit).
ShadowBroker_v0.9.81.zip 6.06 MB
af8c87ccdece8fbb9aadc6be63cce10d3fcba74e6d87ef83289dda6d555fd270
ShadowBroker_0.9.81_x64_en-US.msi 122.4 MB
8977c9a1c54e1f0d030436be9c4e3d81d766cc0080699eb747649095f360c7ff
ShadowBroker_0.9.81_x64-setup.exe 76.5 MB
4e866fa0423c0c2470ed32f4809167a7815dc23ee7762b69e95681c1f3a28250
Post-merge plan
---------------
Force-move the v0.9.81 tag to this commit and replace ALL release
assets on the GitHub release: zip, msi, exe, both .sig files,
latest.json, SHA256SUMS.txt, release-manifest.json.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
build-frontend-export.cjs stages a desktop-only frontend export tree and
strips the ``force-dynamic`` + ``revalidate`` directives from
``frontend/src/app/layout.tsx`` so Next's ``output: "export"`` can
prerender every route.
The strip regexes only matched LF (``\n``). Any Windows checkout without
``core.autocrlf=input`` has CRLF line endings, the strip silently
no-op'd, and the desktop build failed at the static-export step:
Error: Page with `dynamic = "force-dynamic"` couldn't be exported.
`output: "export"` requires all pages be renderable statically
because there is no runtime server to dynamically render routes
in this output format.
Export encountered an error on /_not-found/page: /_not-found
Reaches every Windows contributor who hasn't normalized line endings
locally. Replacing each ``\n`` in the strip regexes with ``\r?\n``
makes the strip CRLF-tolerant; LF behavior is unchanged.
Verified by running both regexes against the actual layout.tsx (302
bytes removed, force-dynamic + revalidate both gone) and against a
synthetic LF input (296 bytes removed, same outcome).
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Shadowbroker