Original error never reaches the log

What goes wrong

A new catch block or output redirect swallows the original error message and stack trace instead of logging or re-throwing it. Whoever reads the log later sees only a secondary symptom ('failed') and cannot find the cause.

Example

A retry wrapper catches a database error and logs 'sync failed' without the exception; the on-call engineer cannot tell whether the cause was a timeout or a bad password.

Not to be confused with

Not deliberate redaction of secrets, not log injection, not a wrong status claim, and not loss of the business record itself — only the diagnostic is lost.

Discovered by

@sks24git GitHub

Field notes · 14

@sks24git2026-11-05

A catch-all around a setup call swallows every error to tolerate one known stand-in. A real break in setup then leaves the feature off for every visitor, with nothing in the console and no failing check.

Never swallow an error you did not expect. Log it, even when the page carries on.
Catch the case you mean, not every case. A broad catch hides the next real bug.

@sks24git2026-10-22

A helper claimed to follow another helper's failure contract but exited with a bare code and no message. Swap one for the other and every health report reads 'exit 2' with the reason gone.

Test the error text, not just the exit code. A contract that names the reason breaks silently when only the code is checked.

@sks24git2026-10-19beat 4 review rounds

A shell group ends in a blanket success so one optional read may fail, and that also hides a failed write of the required part. The cleanup that follows then deletes the only full copy of the log.

Mask only the step that is allowed to fail. A blanket success hides the failure you needed to see.
Delete the source only after the copy is proven whole.

@sks24git2026-10-09

A probe tried two network protocols in turn and kept only the last error. A wrong password on the first protocol showed up as a timeout from the second, so the real cause never reached the operator.

Report every attempt's failure, not the last. A fallback that hides the first error hides the common mistake.
Test the wire, not the stub. A stub cannot see which protocol was spoken.

@sks24git2026-09-11

A one-liner read a file that did not exist yet, silencing stderr after the input redirect. Redirections apply left to right, so the open fails first and the error prints on every tick, burying the real failure.

Silence stderr before the input redirect. The shell reports a failed open before the later one.
Assert empty stderr in shell tests. Green stdout says nothing about the noise beside it.

@sks24git2026-09-10

Impossible input was clamped into range at ingress, so the stored row became legal and indistinguishable from a healthy one. The query that used to enumerate the anomaly returns nothing, and only a log line is left.

Normalizing bad input deletes the evidence, so count it. The row is repaired; the upstream bug goes invisible.
Match the checks beside it. Log-only next to log-plus-metric is a hole.

@sks24git2026-09-10

A CI path filter listed a suite's test file but not the production file that suite pins. Change only that production code and the job is skipped: the one test that would fail never runs, and the merge is green.

Filter CI on the code under test, not only on the test file. A suite wired to itself never runs on the change it guards.
Ask which edit would skip this job. The answer names the missing entry.

@sks24git2026-09-10

Which module runs in which leg was written twice: once in the job matrix, once in a case statement. Drop a matrix row and every surviving leg decides the module belongs elsewhere, so it runs nowhere and all legs pass.

Two lists that must agree need a check reading both. Neither half can notice the other is gone.
Assert what ran, not that nothing failed.

@sks24git2026-09-10

A typo in the flag that pins the tool produced exactly the same skip line as the legitimate no-op case, and the self-test only ever ran dry. Every guard stays green while the resource leaks back into a shared pool.

A skip line must say which skip it is. One message for two outcomes hides the failing one.
A dry-run self-test proves the parser, never the effect.

@sks24git2026-09-10

An upload of test reports was moved off always-run together with the build steps. The only cancel this job can suffer is its own timeout — the undiagnosed hang the file documents — whose evidence now stops being kept.

Keep evidence collection on the failure you have not diagnosed. Skipping the upload on cancel deletes the only forensics you had.

@sks24git2026-09-10beat 1 review round

A self-test reported its negative control as passing even when the substitution it depends on matched nothing. The check then says 'the guard works' on a file it never changed.

Make a no-op mutation a failure, not a pass. A control that cannot fail is decoration.

@sks24git2026-09-09

A client ask that threw was caught and logged, and the catch also skipped the ref fallback that ran after it. Files the client was asked for were lost entirely, though the ref held them.

A catch decides what happens next, not only what gets logged. Trace the path after the catch as carefully as the happy one.

@sks24git2026-09-07beat 2 review rounds

A rule that hid a floating control under a popover was written for phones, where they collide, but shipped with no width limit. On desktop a 250 ms hover over an unrelated widget blanks the control in an empty corner.

Scope a workaround to the case that needs it. A remedy wider than its problem is a new bug on every other screen.

@sks24git2026-09-01

A helper set its data-loss flag correctly under test, but no test drove that flag through the caller that gates scoring on it. Delete three of the four loss conditions and the whole suite still passes.

Test the wiring, not just the helper. A flag proven set but never proven read guards nothing.
Make the negative path run. A loss test that loses nothing is green for the wrong reason.