Incorrect deobfuscated stack traces about error-tracker

[dreamofabear]

One example: ampproject/amphtml#16069 (comment)

[erwinmombay]

i believe this is a bug in closure compilers sourcemap (think justin mentioned it a while ago)

ill take the issue as part of upgrading closure

[jridgewell]

Yup, the issue here is CC is producing bad sourcemaps that don't correctly map back to the original code.

[dreamofabear]

Couldn't find anything obvious in their issues queue or in g/js-compiler-users. It'd be great if we can follow up with their team with a repro case.

[dreamofabear]

Uno mas: ampproject/amphtml#16218 (comment)

[dreamofabear]

Manually deobfuscating traces will be terrible when public property renaming is enabled with 1-pass.

Malte suggests that this is caused by our post-compile transforms e.g. prepending AMP_CONFIG, replacing the RTV string (AMP_CONFIG change might be fixable by adding \n afterwards). Kris suggested using https://github.com/Rich-Harris/magic-string.

[jridgewell]

Oh man, I hadn't even thought about the AMP_CONFIG being prepended.

[cramforce]

I submitted one fix, but there are at least two more issues to address in ampproject/amphtml#17061

[jridgewell]

ampproject/amphtml#17061 is now fixed. Need to wait 3 weeks for the new cuts to include the corrected source maps.

We'll be able to deploy #55 once ampproject/amphtml#17062 hits production next week.

[dreamofabear]

Looks like ampproject/amphtml#17061 was insufficient and this requires more investigation.