feat(tts-native): platform-native TTS backends for macOS, Linux, Windows#487
Open
muunkky wants to merge 6 commits intoPeonPing:mainfrom
Open
feat(tts-native): platform-native TTS backends for macOS, Linux, Windows#487muunkky wants to merge 6 commits intoPeonPing:mainfrom
muunkky wants to merge 6 commits intoPeonPing:mainfrom
Conversation
Adds scripts/tts-native.sh (Unix) and scripts/tts-native.ps1 (Windows SAPI5) so 'tts.enabled: true' produces actual audible speech. Before this, the integration layer landed in PeonPing#442 was calling backend scripts that didn't exist yet — the hook silently no-op'd. scripts/tts-native.sh branches on uname: - Darwin -> 'say' with rate conversion (rate * 200 wpm) - Linux -> piper preferred when binary + model both present, else espeak-ng; silent exit 0 if neither installed - MINGW* -> delegates to scripts/tts-native.ps1 via powershell.exe - other -> silent exit 0 (debug-logs under PEON_DEBUG=1) Rate / volume passed via 'awk -v' variables to block injection from hostile config. Always exits 0 so TTS failure never fails the hook. scripts/tts-native.ps1 uses System.Speech.Synthesis (SAPI5, not WinRT): - begin/process/end blocks accumulate stdin for ValueFromPipeline input - Rate maps to SAPI -10..+10 via [int][math]::Round((Rate-1.0)*10) - Vol maps to SAPI 0..100 via [int][math]::Round(Vol*100) - Voice case-insensitive match against GetInstalledVoices() - --list-voices / -ListVoices enumerates installed voices per platform - try/catch around synthesis, always exits 0 Tests: - tests/tts-native.bats: 42 scenarios (platform branching, engine priority, unit conversions, stdin handling including shell metacharacters, --list-voices per platform, piper sidecar edge cases) - tests/tts-native.Tests.ps1: 40 Pester scenarios (rate/volume mapping, stdin pipeline binding, voice selection + case-insensitivity, spaced voice names, error containment, -ListVoices) - tests/adapters-windows.Tests.ps1: structural assertions for tts-native.ps1 (syntax, param shape, help header, clamping, no ExecutionPolicy Bypass) and for the install.ps1 copy block Installer wiring: - install.sh: remote-install curl for tts-native.sh + chmod +x (local install already picks it up via scripts/*.sh glob) - install.ps1: new copy-or-download block matching the existing win-play.ps1 / win-notify.ps1 pattern Also fixes the peon CLI shims to prefer pwsh (PowerShell 7+) with a fallback to powershell.exe. Environments with PS 7 installed can end up with PSModulePath leaking PS 7 module dirs in front of the 5.1 inbox paths; PS 5.1 then tries to load PS 7's incompatible Microsoft.PowerShell.Security module and fails to resolve Get-ExecutionPolicy, breaking every invocation through peon.cmd and the bash 'peon' wrapper. pwsh uses its own isolated paths; 5.1-only users see identical behavior. Covered by 2 structural tests in adapters-windows.Tests.ps1.
|
@muunkky is attempting to deploy a commit to the Gary Sheng's projects Team on Vercel. A member of the Team first needs to authorize it. |
The /usr/bin/python3 -> $PEON_PY portability fix from the TTS sprint (card w3ciyq) got dropped during the squash onto upstream-pr/tts-native, causing tts-native.bats test 815 to fail on CI. Restores the PATH-based python3 resolver at file load time and swaps all five /usr/bin/python3 callsites inside run_peon_tts() and enable_debug_logging() to use "$PEON_PY".
Five pre-existing Pester failures on main (and PR 487) from PR PeonPing#475 trailed the install.ps1 refactor that replaced raw `-like` matching with the Test-PathRuleMatch helper. - adapters-windows.Tests.ps1 (3x "bindings marks active rule with asterisk"): regex updated from '\$marker.*-like.*\$rule\.pattern' to match the Test-PathRuleMatch call site. - adapters-windows.Tests.ps1 ("evaluates path_rules against event cwd"): regex updated from 'cwd.*-like.*\$pattern' to the Test-PathRuleMatch call form. - peon-packs.Tests.ps1 ("ide_rules evaluation runs after path_rules and before rotation"): IndexOf('\$config.pack_rotation') was matching the earlier '\$config.pack_rotation_mode' substring, making rotIdx smaller than ideIdx. Anchored on 'elseif (\$config.pack_rotation' which only appears at the rotation fallback site. The 6th CI-only failure ("status --verbose shows IDE rule and excluded path context") passes locally on Windows but fails on GH Windows runner, likely due to short-path (RUNNER~1) vs long-path resolution in Test-PathRuleMatch when comparing \$PWD.Path to a config value. Left for a separate investigation.
…unner The "status --verbose shows IDE rule and excluded path context" test was CI-only-red (passed locally). Root cause: on GitHub Windows runners \$env:TEMP resolves to the short-name form (C:\Users\RUNNER~1\...), so the config stored the short form while Set-Location + \$PWD.Path inside the spawned peon.ps1 shell returned the long form. Test-PathRuleMatch compared them as strings and missed the match, so the expected "path rules skipped here (exclude_dirs): ..." line never printed. Resolve-Path both WorktreesDir and ProjectDir to their canonical long-path form after creating them, so both sides of the comparison stay in sync regardless of how \$env:TEMP is spelled.
Adds an inline probe that echoes $PWD.Path and the configured exclude pattern from the spawned shell, and emits the full captured output as a -Because message on assertion failure. Purely diagnostic; does not change the assertion semantics. Will be reverted once the root cause of the CI-only failure is understood.
…ose test Diagnostic probe on the GH Windows runner confirmed the mismatch: PWD.Path = C:\Users\runneradmin\AppData\Local\Temp\... (long) ExcludePattern = C:\Users\RUNNER~1\AppData\Local\Temp\... (short) On GitHub Windows runners, \$env:TEMP resolves to the short 8.3 form (C:\Users\RUNNER~1\...), and Resolve-Path preserves that form. But Set-Location + \$PWD.Path in the spawned shell expands the short name to the long form — so Test-PathRuleMatch compared short-form pattern against long-form PWD and missed. Use Push-Location + \$PWD.Path on both WorktreesDir and ProjectDir at test-setup time so the config values match the form peon.ps1 will see. Diagnostic Write-Host probe from the prior commit is removed now that the root cause is understood.
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
Ships
scripts/tts-native.sh(Unix) andscripts/tts-native.ps1(Windows SAPI5) sotts.enabled: trueproduces audible speech on every supported platform. Before this, the integration layer landed in #442 was calling backend scripts that didn't exist yet — the hook silently no-op'd.Built and tested against the interfaces already committed in ADR-001 (
docs/adr/ADR-001-tts-backend-architecture.md, landed in v2.20.1) and against the existingResolve-TtsBackend/Invoke-TtsSpeakpaths inpeon.ps1andpeon.sh.What's in the PR (7 files, +1736 / -5)
New files
scripts/tts-native.sh— Unix backend. Platform-branches onuname: Darwin usessay, Linux preferspiperwhen binary+model both present elseespeak-ng, MINGW/MSYS2 delegates totts-native.ps1viapowershell.exe. Rate / volume passed viaawk -vvariables to block injection from hostile config. Always exits 0.scripts/tts-native.ps1— Windows SAPI5 backend viaSystem.Speech.Synthesis.begin/process/endblocks for pipeline-bound$InputText. SAPI rate mapping[int][math]::Round(($Rate-1.0)*10)clamped to-10..+10. Volume mapping[int][math]::Round($Vol*100)clamped to0..100. Voice resolution viaGetInstalledVoices()(case-insensitive).-ListVoicesswitch emits installed voice names. try/catch around synthesis, always exits 0.tests/tts-native.bats— 42 BATS scenarios: platform branching (Darwin/Linux/MINGW/unknown), engine priority (piper vs espeak-ng), unit conversions (rate and volume across all three engines),--list-voicesper platform, contract (empty stdin, shell metacharacter safety, missing positional defaults), piper sidecar edge cases.tests/tts-native.Tests.ps1— 40 Pester scenarios: rate/volume mapping and clamping, stdin pipeline binding (production path), voice selection + case-insensitivity, SAPI5 spaced voice names (e.g.Microsoft David Desktop),-ListVoicesoutput, error containment.Modified
install.sh— 2 lines: remote-installcurlfortts-native.sh+chmod +x. Local install already picks it up via the existingscripts/*.shglob.install.ps1— new copy-or-download block fortts-native.ps1matching the existingwin-play.ps1/win-notify.ps1pattern. Also adds pwsh-with-fallback to the generatedpeon.cmdandpeonbash shims (see below).tests/adapters-windows.Tests.ps1— structural tests fortts-native.ps1(syntax, param shape, comment-based help, clamping, noExecutionPolicy Bypass),install.ps1copy-block assertion, and 2 tests for the pwsh-fallback shims.Rides along:
peonCLI shim resiliencyThe generated
peon.cmdand bashpeonwrapper used to hardcodepowershell -NoProfile -NonInteractive .... On environments wherePSModulePathhas PS 7 module dirs ahead of the 5.1 inbox paths (seen on dev boxes with CloudSDK or similar), PS 5.1 tries to load PS 7's incompatibleMicrosoft.PowerShell.Securitymodule and fails withGet-ExecutionPolicy : module could not be loaded, breaking everypeon ...invocation.Both shims now probe for
pwshfirst (where pwshin cmd,command -v pwshin bash) and fall back topowershell.exeonly when pwsh isn't on PATH. PS 5.1-only users see identical behavior. Covered by 2 structural tests.I kept this in the same commit because the shim fix surfaced while smoke-testing
tts-native.ps1through the fullpeonCLI pipeline. Happy to split into a separate commit if you'd prefer that shape.What's NOT in the PR (consciously held back)
VERSIONorCHANGELOG.mdbump — that's your call at release time..gitban/changes.tests/peon-engine.Tests.ps1, hardcoded/usr/bin/python3intests/setup.bash) are held for separate tiny PRs if you want them.Test plan
tests/tts-native.bats42/42 green locally (Git Bash + bats-core)tests/tts-native.Tests.ps140/40 greentests/adapters-windows.Tests.ps122/22 green on the new/modified tests (no regressions expected on the other 416)'hello' | pwsh -File scripts/tts-native.ps1)say) and a Linux host withespeak-ng('hello' | bash scripts/tts-native.sh default 1.0 0.5)tts.enabled: falsebaseline (measure via[exit] duration_mslog line)Notes for reviewers
say,piper,aplay,espeak-ng,powershell.exe) via PATH-first stubs that log invocation shape to files. The real script body runs against stubs; noinstall_mock_tts_backendchange.PEON_TTS_DRY_RUN+PEON_TTS_TRACE_FILEmechanism ontts-native.ps1so behavior can be asserted without actually speaking. Design-sanctioned testability hook.