Internal errors in third pass go uncaught

[gnprice]

We have a nice report_internal_error mechanism for catching unexpected exceptions and reporting them explicitly as internal errors with the source file and line number that triggered the issue.

In the crash reported as #1319, that mechanism didn't fire. That was in the third pass of semantic analysis, and from the code it looks like we invoke that mechanism around the main (second) pass of semantic analysis and the type checker itself, but not the third pass.

Because it seems somewhat tricky to get this error reporting into every context where we know a specific line number, it might be helpful, in addition to any fix specifically for ThirdPass, to put a backstop layer of reporting that makes sure we at least identify the file. (As #1319 illustrates, it can be tricky to identify that from the outside!) It looks like the State.process call in the main build loop in BuildManager.process could be an effective spot for that.

[JukkaL]

A good idea!

[gvanrossum]

Doing this to State.process() would require me to rewrite the logic once incremental lands, so I propose to wait for that (it's also targeted for 0.3.2).

[gnprice]

Sounds good. And thanks for the fast fix!

On Thu, Apr 7, 2016 at 11:15 AM, Guido van Rossum [email protected]
wrote:

Doing this to State.process() would require me to rewrite the logic once
incremental lands, so I propose to wait for that (it's also targeted for
0.3.2).

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
#1341 (comment)