Change Details

Since introduction, configd will report exceptions other than ConfigError as type R_ERROR_DAEMON when running config scripts; this is incorrect --- they should be reported to the shim as commit errors and the error output returned. When running without configd, an exception that is not caught by a config script will trigger a traceback and the airbag report, resulting in a commit error; after the PR below, running under configd will return the error and trigger a commit error. This is observed in the context of (the first fix in) T6592 and will aid in debugging, now that current runs all scripts through configd (T6510). Note that this is unrelated to the formatting issues of T6605. Moreover, this is a case where formatting //should// differ between running with/without configd support: errors uncaught by a config script should be clear for debugging and resolution, and (ideally) not make it the point of the user-facing airbag report.

Since introduction, configd will report exceptions other than ConfigError as type R_ERROR_DAEMON when running config scripts: this has the effect of indicating 'pass-through' execution to the shim, which will re-execute the config mode script within the config session context. The repurposing of the error name is misleading in this case, but the design is reasonable when the config daemon was still operating with a mixture of scripts running natively and under the daemon: if not a clearly actionable ConfigError, return to the native context. With all scripts now running under the config daemon (T6510), one finds that this behavior can obscure some errors. This is observed, for example, in the first fix of T6592: running without the config daemon, the error is triggered on the first call of 'commit', but not a subsequent one, hence the pattern above will not raise an error. Since EAFP style within a config script is never exhaustive, unexpected errors should be caught and reported by the config daemon and a commit error code returned to the shim. This provides a cleaner version of what happens when running without configd: namely, an exception that is not caught by a config script will trigger a traceback and the airbag report, effectively resulting in a commit error. Note that this is unrelated to the formatting issues of T6605. Moreover, this is a case where formatting //should// differ between running with/without configd support: errors uncaught by a config script should be clear for debugging and resolution, and (ideally) not make it the point of the user-facing airbag report. In testing, this refinement has revealed several other obscured bugs; those will be itemized in subtasks for reference.

Since introduction, configd will report exceptions other than ConfigError as type R_ERROR_DAEMON when running config scripts;: this has the effect of indicating 'pass-through' execution to the shim, this is incorrect --- they should be reported to the shim as commit errors andwhich will re-execute the config mode script within the error output returnedconfig session context. When runnThe repurposing without configdof the error name is misleading in this case, an exception that is not caught by a config script will trigger a traceback and the airbag reportbut the design is reasonable when the config daemon was still operating with a mixture of scripts running natively and under the daemon: if not a clearly actionable ConfigError, resulting in a commit error;return to the native context. aftWith all scripts now running under the PR belowconfig daemon (T6510), running under configd will return the error and trigger a commitone finds that this behavior can obscure some errors. This is observed in the context of (the first fix in) T6592 and will aid in debugging, for example, in the first fix of T6592: running without the config daemon, the error is triggered on the first call of 'commit', but not a subsequent one, hence the pattern above will not raise an error. Since EAFP style within a config script is never exhaustive, unexpected errors should be caught and reported by the config daemon and a commit error code returned to the shim. This provides a cleaner version of what happens when running without configd: namely, an exception that is not caught by a config script will trigger a traceback and the airbag report, now that current runs all scripts through configd (T6510)effectively resulting in a commit error. Note that this is unrelated to the formatting issues of T6605. Moreover, this is a case where formatting //should// differ between running with/without configd support: errors uncaught by a config script should be clear for debugging and resolution, and (ideally) not make it the point of the user-facing airbag report. In testing, this refinement has revealed several other obscured bugs; those will be itemized in subtasks for reference.