Compare commits
merge into: justin/webhook-server:gitea-sync
..
pull from: justin/webhook-server:claude/pensive-easley-4abcbe
10 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
|
|
b93759ff8a |
build-installer: run ISCC from the .iss directory (fixes Gitea runner)
Some ISCC builds reject deeply-nested absolute .iss paths (the runner runs out of %SystemRoot%\System32\config\systemprofile\.cache\act\...) with a misleading 'The system cannot find the path specified.' before touching any source files. Push-Location to the .iss directory and pass the bare filename - same effect, no path quirk. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
|
f83d194002 |
Installer build: pre-flight diagnostics + verbose ISCC output (#9)
ISCC's default error message ("The system cannot find the path
specified.") doesn't say which file failed, which made the Gitea
runner failure undebugable. Two changes:
1. Before invoking ISCC, log a one-line summary per source path the
.iss references: file count, longest path length, root. Also dump
HKLM\...\FileSystem\LongPathsEnabled so MAX_PATH issues are
immediately obvious.
2. Pass /Qp to ISCC so it prints each file as it processes; the line
right before the error tells us which source file ISCC choked on.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
|
1b5827b8f4 |
v0.1.4: Rollback NRE fix + Gitea Actions builds (#8)
Release (Gitea) / build-installer (push) Failing after 26s
* Fix Rollback NRE: capture filename before refreshing the checkpoint list The actual restore on the service was succeeding but the GUI showed "Rollback failed: Object reference not set to an instance of an object." because RefreshAsync clears the ObservableCollection, the SelectedItem binding goes null when its source item disappears, and the next line dereferenced Selected!.FileName. Cache the filename and timestamp before any await so the post-refresh status update doesn't depend on Selected still being non-null. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * v0.1.4 --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
|
68706df2c5 |
Gitea Actions: workflow guards + .gitea release pipeline (#7)
* CI: skip GitHub-specific workflows on Gitea Actions; add workflow_dispatch Gitea reads .github/workflows/ for compatibility, so without guards it tries to run the release and wiki-sync workflows on Gitea too - which fail because they use softprops/action-gh-release@v2 (GitHub-only) and push to GitHub-hosted wiki URLs. Add `if: github.server_url == 'https://github.com'` to the release-builder and wiki-sync jobs so Gitea no-ops them. ci.yml (plain dotnet build + test) still runs on both, and now also accepts workflow_dispatch so it can be triggered by hand to verify a self- hosted runner is picking up jobs. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Gitea release workflow: build installer + attach to Gitea release .gitea/workflows/release.yml runs on the self-hosted Windows runner when a v* tag is pushed (or via workflow_dispatch with a version override for testing without bumping the project). Mirrors the GitHub release workflow shape - dotnet test, choco-install Inno Setup if missing, run scripts/build-installer.ps1 - then uses Gitea's REST API directly for the release-creation + asset-upload steps (Gitea has no equivalent to softprops/action-gh-release). Token is the runner-injected secrets.GITHUB_TOKEN, which Gitea provides for GitHub Actions compatibility. Path location matters: lives under .gitea/ so GitHub Actions never sees it, while the existing .github/workflows/release.yml stays GitHub-only (already gated with `if: github.server_url == ...`). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
|
0b7c20a1fa |
v0.1.3: hide-to-tray + tray toggle + context menu fix (#6)
* GUI: hide-to-tray on X button; tray persists until explicit Exit The minimize-to-tray behavior already worked, but clicking the X button killed the GUI process and took the tray with it. That made "tray when the GUI window is closed" a UX dead end - the only way to get the tray was to leave the window minimized. Now: - X button / Alt+F4 -> hide window, tray stays alive - Tray double-click -> reopens window - File -> Exit (or tray's Exit menu) -> truly quits the process Wired by adding a RealExitRequested event on MainViewModel that the window subscribes to (so File -> Exit sets the ExitForReal flag before calling Shutdown), and a parallel onExit callback on TrayIcon for the tray menu's Exit item. The Closing handler checks ExitForReal: if false (X / Alt+F4) it cancels the close and hides; if true, it disposes the tray and lets the close proceed. Auto-start at login is still TBD - if you want the tray to be there without manually launching the GUI after a reboot, that's a separate Task Scheduler entry. Skipping for now. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Add File -> Minimize to tray toggle (default on) Adds a checkable MenuItem so the user can opt out of the hide-to-tray behavior. Persisted per-user to %APPDATA%\WebhookServer\gui.json so the choice survives restarts. When ticked (default): X / Alt+F4 / minimize hide to tray, GUI process keeps running, tray icon persists. When unticked: X actually closes the app, minimize is a regular Windows minimize. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Fix endpoint-row context menu: bindings via PlacementTarget.Tag The ContextMenu lived in its own popup visual tree, so the menu items' RelativeSource={RelativeSource AncestorType=Window} couldn't find the Window and the bindings silently failed - none of Edit / Copy URL / Toggle / Delete actually fired their commands. Standard WPF workaround: park MainViewModel on each DataGridRow's Tag (still in the Window's visual tree, so the row Setter binding resolves) and reach it from the menu items via PlacementTarget.Tag. The toggle command parameter likewise comes from PlacementTarget.DataContext (the EndpointConfig the row represents). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * v0.1.3 --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
|
9fcff2694a |
Wiki sync: stop treating git's stderr warnings as fatal (#5)
PowerShell with ErrorActionPreference=Stop escalates ANY native-command stderr output to a script-terminating error. git writes plenty of informational lines to stderr (CRLF nags, "remote: Processed N references", "Switched to branch X"), which made the sync script abort partway through every run when actually nothing was wrong. Three fixes: 1. Switch to ErrorActionPreference=Continue and check $LASTEXITCODE manually after each git call. 2. Drain stderr on each git invocation with `2>&1 | Out-Null`. 3. Disable core.autocrlf and core.safecrlf in the throwaway wiki clone so git stops complaining about line endings. Verified end-to-end against Gitea: 12 pages + sidebar pushed cleanly. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
|
8e514f29fc |
Add wiki sync: docs/ stays the source of truth, wikis auto-mirror (#4)
scripts/sync-wiki.ps1 clones a wiki repo, copies+flattens markdown from docs/ with a slug mapping (e.g. recipes/zerto-pre-post-scripts.md becomes the Recipe-Zerto-Failover page), rewrites in-repo markdown links to wiki-style targets, generates a _Sidebar.md, and pushes back if anything changed. Idempotent. .github/workflows/wiki-sync.yml runs the sync on every push to main that touches docs/ (or the sync tooling itself). Uses GITHUB_TOKEN which has wiki write access via the contents:write permission. For Gitea, no Windows runner is available, so the script is invoked manually with a Gitea PAT in the URL. One-time setup for each remote: enable Wiki in repo settings, create a Home page via the web UI to initialize the wiki repo, then run the sync. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
|
f00ee0cf3a |
v0.1.2: Config Checkpoints dialog, descriptions, daily auto-snapshot, docs (#3)
* Documentation: install/upgrade/uninstall guides + recipes incl. Zerto Adds a docs/ folder under the repo root with full operator documentation aimed at sysadmins (not webhook developers). The Zerto pre/post script recipe is the canonical "why does this exist" walkthrough; the GitHub HMAC, AD password reset, and UI-on-desktop recipes round out common patterns. Pages: - README.md (index) - concepts.md (5-minute "what is a webhook" explainer) - installation.md (interactive + silent install) - upgrading.md (single-click upgrade flow + edge cases) - uninstalling.md (clean removal + wiping ProgramData) - runas-modes.md (Service / InteractiveUser / SpecificUser decision flow) - service-account-and-ad.md (gMSA setup, delegated rights) - network-and-security.md (bind addresses, allowlists, HTTPS, secret storage) - troubleshooting.md (symptom -> first check, common errors) - recipes/zerto-pre-post-scripts.md (canonical use case) - recipes/github-style-hmac.md (GitHub / Stripe-shaped webhooks) - recipes/ad-password-reset.md (gMSA-backed self-service reset) - recipes/ui-on-desktop.md (InteractiveUser pattern) Top-level README.md restructured to point at docs/ as the source of truth, dropping the duplicated installation snippets. Installer ships docs/ alongside the binaries so they're available offline at C:\Program Files\WebhookServer\docs\. GUI Help menu gains a "Documentation" item that opens the docs site in a browser. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Config Checkpoints dialog + daily auto-checkpoint; drop installer GUI launch Three fixes: 1. Config Checkpoints submenu replaced with a proper dialog. Lists checkpoints with timestamp/size/filename, has a "Take Checkpoint Now" button, and a "Roll Back" button that becomes enabled when a row is selected. The previous click-a-menu-entry-immediate-restore flow was too easy to fire by accident. 2. New CheckpointScheduler BackgroundService creates a checkpoint at midnight every day. Combined with the existing auto-on-save snapshots, this guarantees a daily rollback point even if the config wasn't edited that day. A new "create-checkpoint" admin op plus AdminPipeServer.CreateCheckpoint helper does the actual file copy; both manual (via the dialog) and the scheduler use it. 3. Installer: drop the post-install "Launch Webhook Server" wizard step. It tried to launch the GUI un-elevated, which fails because the GUI's manifest is requireAdministrator. The Start Menu shortcut handles elevation correctly, so the user can launch from there. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Docs: replace AD-reset recipe with realistic Zerto failover walkthrough The AD password reset endpoint was a poor fit for what people actually need this server for. Replaced with a realistic Zerto post-failover example that's much closer to the project's purpose: - Update DNS A records for failed-over hostnames - Wait for the VM to come up at the DR site - PowerShell-remote into the VM and check / start critical services - Notify Teams with the result The flagship pattern is now: Zerto post-script (curl, fire-and-forget) calls an Async webhook endpoint -> 202 in milliseconds -> Zerto's failover sequence is never blocked. The server runs the actual work in the background, with full output captured in the daily log. A ready-to-use Zerto-side script ships at scripts/examples/zerto-post-failover.ps1 - pure curl.exe (no PowerShell modules), reads the bearer token from a file the ZVM service account can read. The installer now bundles scripts/examples/ alongside docs/ so the example is also available locally at C:\Program Files\WebhookServer\scripts\examples\. Removed: docs/recipes/ad-password-reset.md. Updated: docs/README.md, README.md, the recipe content itself. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Restore installer GUI launch (via shellexec) + checkpoint descriptions Two follow-ups to the previous Config Checkpoints commit: 1. Bring back the post-install "Launch Webhook Server" checkbox in the installer. The previous attempt failed because Inno Setup's postinstall flag launches via CreateProcess after Setup exits, bypassing the GUI's requireAdministrator manifest. Adding the shellexec flag switches to ShellExecute, which DOES honor the manifest and triggers a clean UAC prompt - so the post-install GUI launch works as expected. 2. Each checkpoint now carries a description, stored in a sidecar .meta.json file next to the snapshot. Defaults: - Auto-on-save: "Before save" - Midnight scheduler: "Nightly auto-checkpoint" - Manual: opens a small dialog so the user can type a meaningful description (defaults to "Manual checkpoint" if blank) The dialog and pruning both clean up sidecars alongside snapshots. The Config Checkpoints grid grows a Description column between When and Size. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * v0.1.2: bump checkpoint retention 30 -> 90 Each checkpoint is a few KB of JSON plus a tiny sidecar; even at 90 entries on a config with hundreds of endpoints the on-disk footprint is negligible (worst case ~20 MB). With daily auto-checkpoints plus on-save snapshots, 30 entries could fill in a couple weeks of moderate use; 90 gives a comfortable ~3-month window. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
|
7d94535d5d |
v0.1.1: GUI auto-elevates, installer handles upgrades cleanly (#2)
* v0.1.1: GUI auto-elevates, installer stops service before file copy Two fixes for the v0.1.0 install experience: 1. Embed app.manifest with requestedExecutionLevel=requireAdministrator so the GUI always elevates. The named pipe is ACL'd to SYSTEM and the Administrators group, but UAC token splitting puts Admins in deny-only on the standard token, so launching the GUI from the Start Menu fails to connect with "Access is denied". The manifest forces UAC to elevate, surfaces the shield icon on the shortcut, and matches the reality that the GUI cannot function without admin rights. 2. Add a [Code] PrepareToInstall hook to webhook-server.iss that runs `sc stop WebhookServer` before file copy. Upgrade installs were failing on locked binaries because the running service held the exes open. sc returns non-zero on fresh installs (no service yet) which we ignore. Bumps Version to 0.1.1. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Rename "Backups" menu item to "Config Checkpoints" User-facing copy only; internal API names (Backups collection, BackupEntry, list-backups op, etc.) stay the same to avoid churn through the wire protocol and existing on-disk files. The new phrasing makes the auto-snapshot-before-save model more discoverable. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Installer: synchronous service stop + kill stray GUI/Service processes The previous sc.exe stop is fire-and-forget; on slower machines the file-copy step started before the service had actually released its binaries, leaving the upgrade in a broken state. Switch to net.exe stop which blocks until the service reports STOPPED. Also taskkill any running WebhookServer.Gui.exe (the user might have left the tray running) and any orphan WebhookServer.Service.exe (from deploy.ps1 dev runs) so all copies of the binaries are unlocked before [Files] runs. Pre-flight ServiceExists() check via sc query so the installer only calls "net stop" when there is actually a service to stop, rather than relying on net's error code. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
|
a808964cf1 |
Phases 1-7: GUI polish, icons, tray, backups, installer, CI (#1)
* Phase 3: app icon (multi-resolution ICO + master PNG) scripts/generate-icons.ps1 renders the icon programmatically with System.Drawing - rounded teal square (#0E7C66) with a stylized white hook glyph - at 16/24/32/48/64/128/256 px and assembles a proper multi-resolution Microsoft ICO. The PNG and ICO outputs land in resources/. The script is the source of truth; re-run after editing the design. GUI csproj uses ApplicationIcon for the EXE icon and embeds the .ico + .png as Resources so MainWindow and AboutDialog can use them via WPF's resource URI scheme. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Phase 5: tray icon with minimize-to-tray and context menu GUI csproj enables UseWindowsForms (NotifyIcon lives in WinForms even in .NET 8). New Services/TrayIcon.cs wraps NotifyIcon with a context menu (Open / Restart service / Exit) and the embedded webhook-server icon. MainWindow creates the TrayIcon, hides itself on minimize and restores on tray double-click. Adds GlobalUsings.cs to alias the WPF defaults for types that exist in both WPF and WinForms (Application, MessageBox, TextBox, Binding, etc.) so existing code keeps compiling without per-file changes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Phase 6+7: Inno Setup installer + GitHub Actions release pipeline installer/webhook-server.iss is an Inno Setup 6 script that: - Installs to %ProgramFiles%\WebhookServer - Creates Start Menu folder + GUI shortcut (and optional desktop icon) - Runs install-service.ps1 post-install to register the Windows Service - Runs uninstall-service.ps1 pre-uninstall to remove it - Bundles the webhook-server icon for the installer / uninstaller scripts/build-installer.ps1 is the local build helper: publishes both projects, finds ISCC.exe (PATH or standard install path), compiles the installer with the version pulled from Directory.Build.props, drops the output in dist/. .github/workflows/ci.yml runs build + test on every push/PR to main. .github/workflows/release.yml triggers on v* tags (or manual dispatch), runs tests, installs Inno Setup via choco, builds the installer, and attaches the .exe to a GitHub Release. Pre-1.0 versions are flagged prerelease automatically. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Phase 4: backups + import/export config ConfigStore.SaveAsync now snapshots the previous config to %ProgramData%\WebhookServer\backups\config-<timestamp>.json before overwriting, retaining the last 30. Failures are silent so a backup-write hiccup never blocks an actual save. Three new admin pipe ops: - list-backups: returns newest 50 entries with timestamps and sizes - restore-backup: takes a fileName, refuses path-traversal chars, loads the named backup over the live config (which itself triggers a fresh backup of the current state via the SaveAsync hook) - import-config: replaces the current config with a GUI-supplied ServerConfig, merging encrypted secrets where the GUI didn't supply new plaintext GUI File menu items are wired: - Import config: file picker -> ImportConfigAsync - Export config: SaveFileDialog writes the current config as JSON - Backups: dynamic submenu auto-refreshed when opened, listing backups with timestamp + size; click to confirm-and-restore Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |