Switch imports to charm.land/bubbletea/v2 and charm.land/lipgloss/v2.
View now returns tea.View with AltScreen set, key handling uses
KeyPressMsg, and colour fields use image/color.
* feat: add --context to forward only selected contexts
kportal forwards every context in the config file. When the file describes
several clusters that host the same services, that means holding connections
open to clusters nobody is using, and - because all contexts are active at once
- every localPort has to be globally unique. The same service therefore lands on
a different local port per cluster, so switching clusters means reconfiguring
the app being developed.
--context limits a run to the contexts you name. It is repeatable and accepts
comma-separated values, and composes with -headless:
kportal --context development-team-a
kportal --context team-a,team-b -headless
kportal --context team-a --context team-b
An unknown name is an error listing the contexts the file defines, so a typo
fails loudly instead of silently forwarding nothing.
Local-port and mDNS-hostname conflicts are now scoped to forwards that could run
at the same time: same context, or both contexts in the active selection. With
no selection every context is active, so existing behaviour is unchanged.
Contexts that are never selected together may reuse the same localPort, which is
what lets a service keep one fixed local port in every cluster.
Validation still covers the whole file, so a problem in a context outside the
selection is still reported and --check does not get weaker. Loading, validating
and filtering are shared by every path that re-reads the config (startup, file
watcher, SIGHUP in both headless and verbose modes), so no path can forget the
selection.
Also fixes a latent identity bug this makes reachable: Forward.ID() is only
"alias:localPort" when an alias is set, which was unique across contexts only
because duplicate local ports were globally rejected. With ports now shareable,
deleting or editing a forward from the TUI could have matched same-alias
forwards in other contexts and silently removed them. Delete and edit are now
scoped to the context the row came from.
* fix(config): keep the running config when no selected context remains
A reload dropped selected contexts that no longer exist in the file.
When all of them vanished, the reload returned an empty config and the
manager stopped every forward after a single rename.
LoadForRuntime now returns an error in that case. The watcher keeps the
previous config, as it does for any other invalid reload.
Refs #80
* fix(ui): apply new table columns after a successful SIGHUP reload
The context selection moved the SIGHUP reload into reloadFromSignal,
which had no way to update the verbose table columns. The function now
takes a callback that runs only after the manager reloads without error.
The verbose table passes SetColumns.
Refs #80
* style: gofmt run_test.go
---------
Co-authored-by: André Cruz <[email protected]>
Some users cannot list namespaces on a cluster. The add wizard showed
an error and offered no way forward.
The namespace step now switches to a text prompt when the list call
fails or returns nothing. The prompt shows the reason and checks that
the name is a valid DNS label. Esc from a later step returns to the
prompt.
The j and k keys now type into text fields. They moved the list cursor
before, so a name such as kube-system could not be typed.
Closes#66
* Allow configuring forwards-table columns, order, and width
Adds an optional tui.columns section to .kportal.yaml so the visible
columns, their order, and their widths can be customized instead of
always rendering the hardcoded 8-column layout. Omitting the section
keeps the existing default columns/order/widths.
Closes#81
* fix(ui): keep the plain table layout and check column widths
The plain verbose table must look the same when no tui section is
set. It had narrower columns and shorter port headers than before. The
default layout is now identical to the old one, and a golden test
guards it.
A status column that is not last now pads by visible width. Table
rules follow the column widths. The UI applies new columns only after a
reload succeeds, including the SIGHUP path of the verbose table.
The validator rejects a column width above 200. The README describes
the tui.columns setting.
Refs #81
* fix(httplog): replace the deprecated ReverseProxy Director with Rewrite
Staticcheck reports SA1019 because Go 1.26 deprecates Director. The
proxy now sets the target in a Rewrite function. Rewrite drops the
X-Forwarded-For header that Director sent, so the function calls
SetXForwarded to keep that header.
A new test checks the target port, the path and the header.
* fix(release): drop the deprecated cask url verified key
Homebrew warns about the verified parameter of the cask url stanza.
The setting is not needed. GoReleaser check now passes without a
deprecation warning.
The postflight warning remains. GoReleaser generates it from the
post install hook, and the upstream fix is not released yet.
Refs #79
---------
Co-authored-by: Lukasz Raczylo <[email protected]>
CI on Linux flagged a race in TestStartWorker_HealthCallback_StatusChange:
the test read ui.updates while the healthchecker's per-port goroutine
(spawned by Register at checker.go:164) was still running and could
fire UpdateStatus through notifyStatusChange.
The earlier mutex on MockStatusUpdater protected the writes; the read
side was unprotected, and the goroutine had not finished by the time
the test started ranging over the slice on slower runners.
Fix:
- call healthChecker.Unregister(fwd.ID()) to drain the per-port
goroutine before reading
- hold ui.mu around the slice read for belt-and-suspenders happens-
before, regardless of goroutine timing
Verified locally with go test -race -count=20 on the targeted test
and -count=3 on the full forward package.
The mock recorded calls into unprotected slices while being invoked
concurrently from the test goroutine and from the health-checker
callback registered by Manager.startWorker (and the watchdog hung
callback). go test -race -count=20 reliably tripped the race detector
on TestStartWorker_HealthCallback_StatusChange.
Wrap the slice writes in a sync.Mutex. Read-side code in tests reads
only after manager.Stop() drains the background goroutines (Stop's
wg.Wait establishes a happens-before edge), so direct reads remain
race-free.
internal/ui: 27.9% -> 79.8%. Adds three test files (~3300 lines):
- wizard_handlers_extended_test.go: keyboard branches not previously
covered (ctrl+c, pgup/pgdn, search filter typing, port-checked
failure path, alias text edit, focus toggling, error states),
plus removeForwardsCmd / removeForwardByIDCmd against real Mutator
+ tempdir config (dedup + at-least-one-namespace validation).
- wizard_views_test.go: every renderXxx path with a populated
AddWizardState — context/namespace selectors, resource selection,
port entry, http log list/detail, success, scroll indicators.
- table_test.go: AddForward / UpdateStatus / SetError / Clear /
FilterStrings flows.
Test-only fixups noted by agent:
- getFilteredEntries silently skips StatusCode==0 entries (only
completed responses are visible). Tests now seed StatusCode=200.
- handleForwardSaved leaves step=StepConfirmation on save error
rather than advancing to StepSuccess; assertion corrected.
- removeForwardsCmd needs >=2 forwards in a context for dedup tests
so removing one leaves the context non-empty (validator rejects
empty contexts).
go test -race -count=1 ./... clean across all 14 packages.
cmd/kportal: 0% -> 17.4% (testable surface only — main() is a
274-statement monolith with OS signals, bubbletea TUI, kubeconfig
loading, signal-loop goroutines, and config watcher all in one
function. Hitting 70% requires extracting it into run(args,w) int,
which is a non-trivial refactor — out of scope for a coverage pass.)
Tests cover: promptCreateConfig (yes/no/EOF/empty inputs via os.Pipe
stdin redirection), contains, resolveGenerateConfigPath (all 4 system
dirs + abs/rel branches), runGenerate (missing flag, -h, unknown
flag, invalid context, malformed YAML, no-TTY error path).
internal/version: 45.7% -> 97.8%. Added httptest.NewServer + custom
rewriteTransport to redirect GitHub API calls. Covered NewChecker,
fetchLatestRelease (200/403/404/429/500, malformed JSON, empty
tag_name), CheckForUpdate (newer/same/current-newer/error/cancelled),
parseVersion edge cases (empty, single digit, alpha).
New subcommand that connects to a chosen kube context and walks the
user through three picker steps before writing forwards to the config:
1. Namespace multi-select (no system-namespace exclusion)
2. Service multi-select grouped by namespace, with already-configured
rows greyed out and locked off
3. Starting-port input with a live preview of consecutive local-port
assignments, skipping any port already taken by the existing config
or another row in the batch
Multi-port services emit one forward per port. Non-TCP ports are
skipped with a one-line warning at the end (kportal forward layer is
TCP-only). --dry-run prints the planned forwards without writing.
Subcommand dispatch lives in cmd/kportal/main.go: when os.Args[1] ==
'generate', runGenerate(os.Args[2:]) is invoked before the main
flag.Parse() so the flag.NewFlagSet 'generate' parses its own
'--context', '--config', '--dry-run' flags. Invalid contexts list
the available ones for quick correction.
Tests in internal/ui/generate_test.go cover:
- namespace toggle / toggle-all / filter
- service multi-select with locked already-configured rows and
non-TCP filtering
- port-collision-aware consecutive assignment using a real Mutator
against a tempdir config
- reject + recover for starting-port < 1024
- dry-run does not invoke the mutator
- end-to-end Update() walk through the three steps
- parse-starting-port boundary table
- port-step view rendering
- ServiceCandidate.Key() determinism
README updated with a 'Generate Forwards from a Cluster' section
describing the flow and the three flags.
Previously, editing a forward and keeping its local port unchanged
failed the wizard's port-availability check: the in-config scan
found the forward's own entry and reported '✗ Port N already
assigned to <self.ID>'. Users had to pick a different port,
edit, then change back.
checkPortCmd now accepts an excludeID. The wizard passes
wizard.originalID when isEditing so the forward being edited is
ignored during the in-config conflict scan. The OS-level port
check is unchanged (still catches actual port collisions).
New regression test: TestCheckPortCmd_ExcludeID_AllowsKeepingOwnPort.
Adds an HTTP-log enable toggle to the wizard's confirmation step so
users can flip httpLog on a forward without editing YAML by hand.
Behaviour:
- 'h' on the confirmation step toggles HTTPLog when not focused on
the alias text input. When focus is on alias, 'h' is treated as
text so users can still type aliases like 'host' or 'http-proxy'.
- The confirmation summary shows '[x] enabled' or '[ ] disabled'.
- New forwards: toggle on -> &HTTPLogSpec{Enabled: true}; off -> nil.
- Edit mode: pre-populates the toggle from the existing forward and
preserves any advanced HTTPLog fields the user had configured in
YAML (logFile, includeHeaders, maxBodySize, filterPath) by copying
the original spec on save. Toggling off discards the advanced
fields (consistent with 'absent in YAML = disabled').
State changes:
- ForwardStatus gains *config.HTTPLogSpec so the wizard can see the
full original spec on edit.
- AddWizardState gains httpLog bool + httpLogOriginal *HTTPLogSpec.
Three new tests:
- TestHandleAddWizardKeys_HToggleHTTPLog
- TestHandleAddWizardKeys_HOnAliasFocusIsTextInput
- TestEditPrefill_PreservesHTTPLog
Real kubeconfig context names commonly contain characters the
validator rejected:
- 'admin@home', '[email protected]' (kubectl rename, EKS
aws-iam-authenticator)
- 'cluster.example.com', 'gke_proj_zone_cluster.prod' (FQDN, GKE)
- 'arn:aws:eks:us-east-1:123:cluster/foo' (EKS ARN)
kubeconfig itself imposes no character restrictions, so requiring
[a-zA-Z0-9_-] only was kportal-specific over-validation that blocked
legitimate users. Widen the allowed set to add '@', '.', ':', '/'.
Names must still start and end with a letter or digit so YAML
specials and leading whitespace remain rejected.
Tests cover the new positive cases and tighten negative coverage
(starts-with-@, ends-with-/, ends-with-dot).
Previously a second Stop() call would panic from a double-close on
eventBus and re-stop the healthChecker/watchdog whose contexts had
already been cancelled. Wrapping the body in sync.Once makes
sequential and concurrent double-Stop safe.
TestManager_Stop_Idempotent covers both paths.
P0 #1 — HTTP traffic logger captured Authorization, Cookie, Set-Cookie,
X-Api-Key, X-Auth-Token, X-Csrf-Token, Proxy-Authorization, X-Access-Token
verbatim into log entries (file 0600 + UI subscribers). Bearer tokens
and session cookies were ending up on disk whenever httpLog.includeHeaders
was enabled.
flattenHeaders now redacts:
- the explicit list above (case-insensitive via http.CanonicalHeaderKey)
- any header name containing 'token', 'secret', 'password', 'apikey'
Header names remain visible; values become [REDACTED].
Redaction is unconditional and on-by-default — no opt-out flag. Users
who want raw headers can use tcpdump.
P0 #6 — Headless mode without -v silently routed both structured and
stdlib logs to io.Discard. A daemon under launchd/systemd had no way to
report errors. Headless now defaults log destination to os.Stderr; -v
controls only the level (debug vs info). TUI-quiet path is preserved.
Tests in internal/httplog/redact_test.go cover all explicit names,
substring patterns, and case variants.
In the remove-wizard's confirming state, pressing Esc was reflexively
calling removeForwardsCmd — i.e. confirming deletion. The on-screen
help text said 'Esc: Cancel'. Reflexive Esc-to-cancel destroyed data.
Esc now sets confirming=false and resets the cursor; deletion
requires Enter on Yes. Non-confirming Esc behavior (exit wizard with
ClearScreen) is unchanged.
Three regression tests added in handlers_test.go.
P0 #2 — currentConfig data race
Manager.currentConfig was written without locking in Start/Reload but
read from the health-checker callback goroutine. All accesses now go
through workersMu (read or write as appropriate).
P0 #3 — Reload kills health checker permanently
Reload's zero-forward branch called m.Stop() which tore down the
health checker, watchdog, and event bus. After that, EnableForward
silently registered callbacks against dead components. Now the branch
stops only the running workers; the supervisory infrastructure stays
alive across config changes.
P0 #4 — rest.Config write-write race
executePortForward was mutating .Dial on the cached *rest.Config
shared by all forwards in the same kube context. Cloning the config
with rest.CopyConfig before mutation isolates per-forward dialers.
P0 #5 — ForwardWorker.Stop() double-close panic
close(w.stopChan) is now wrapped in sync.Once, so concurrent Stop
calls (Manager.Stop racing stopWorkerInternal) are safe.
New tests in internal/forward/concurrency_test.go exercise each fix
under -race: 16 concurrent worker Stops, repeated sequential Stops,
empty-Reload preserves infra pointers, and concurrent currentConfig
read/write.
CI runners have no kubeconfig, so clientcmd's loader returns an empty
config (no error) and CurrentContext == "". The previous assertion
'NotEmpty(context)' on the success branch was incorrect — an empty
current-context is valid for an empty kubeconfig.
Mirrors the looser pattern in TestDiscovery_ListContexts.
- [x] Add golangci-lint v2 configuration with formatters section
- [x] Reorganize linters-settings under linters section
- [x] Replace if-else chains with switch statements for clarity
- [x] Wrap all ignored error returns with `_ = ` pattern
- [x] Add OSC 8 hyperlink helper function for clickable ports
- [x] Add blank line in table styling function
- [x] Remove unnecessary type assertion in test
When user starts kportal for the first time, and there is no config file,
kportal should create an empty config file with default values and empty
forwarding rules, so that user can easily edit the config file and add their
own rules.
When adding a service via the wizard, resolve the service's targetPort to the actual pod container port instead of using the service port directly.
Problem: Service port 80 → Pod port 8000, but kportal was trying to forward to port 80 on the pod.
Solution: Look up the pod's actual containerPort when the service uses a named targetPort (like http), and use that for port-forwarding.
* Fix enter misbehaving.
* Cleanup after previous tui implementation.
* Fix race condition and improve logging
* Add filtering of the namespaces by text input in the wizard UI
lukaszraczylo
