1284975 - [meta] Make SpiderMonkey clean on UBSan

Status: NEW

(Core :: JavaScript Engine, defect, P2)

Description

[Terrence Cole [:terrence]]

We have a number of open bugs (e.g. 1281168, 1262015, and everything [gcc 6.1]) that reek of "miscompilation". Fixing all of our UBSan reports seems like it might be a faster way to fix these than hunting down each such miscompilation one at a time.

Luckily, running jsapi-tests under UBSan turned up a number of scary, but correct and actionable, reports that we should fix. I'll file bugs blocking this one with patches. Help would of course be appreciated, if anyone has spare cycles. It's really easy to get UBSan running and it's reports are generally pretty obviously correct and relatively easy to address; ping me on IRC if you need help.

Comment 1

[Nick Fitzgerald [:fitzgen] [⏰PST; UTC-8]]

This is a most noble goal to attain.

Comment 2

[André Bargull [:anba]]

Attachment: ubsan-20170920-5cd2ba3bc6c4.txt

List of UBSan errors when running jstests+jit-tests with default options. Already excluding the issues fixed in bug 1401145, bug 1401191, bug 1401200, bug 1401209, and bug 1401515.

Comment 3

[demiobenour]

Could we build with `-f[wt]rapv -fno-strict-aliasing -fno-delete-null-pointer-checks` and insert a compiler plugin that modifies the IR to not require pointer alignment?

Comment 4

[Takanori MATSUURA]

All depending bugs are now fixed.