Thread (19 messages) flat view 19 messages, 5 authors, 2022-08-19

Re: What's cooking in git.git (Oct 2021, #02; Wed, 6)

From: Jeff King <hidden>
Date: 2021-10-08 21:32:42

On Fri, Oct 08, 2021 at 09:51:33AM +0200, Johannes Schindelin wrote:
quoted
Sort of. They basically wrap the "make" invocation to intercept "cc". My
understanding is that their faux-compiler is mostly about gathering data
about the code. That gets stuffed into a tarball and uploaded to their
servers, where the real analysis happens.

It's very black-box, which I don't love. But in my experience it
produces by far the most useful static-analysis output of any tool I've
seen.
It is pretty black box, but I have to disagree that the static analysis
output is very useful. The majority are false positives about
strbuf/strvec type usage of a static, fixed-size array that is dynamically
replaced by a dynamically-allocated array. Coverity misses that subtlety
and reports out-of-bounds accesses.
Yes, I remember skipping past quite a few of those.

To be clear, I don't claim that its output is amazing. Only that it has
produced actionable output on many occasions. Grepping commit messages
for "Coverity" turns up several hits (many from you :) ). Most of those
are leak fixes, and I do think we have better options there. But I
recall it detecting some hard-to-find memory and logic errors, too.
Granted, I worked around those (I thought) by using the
`-DFLEX_ARRAY=65536` trick, but I guess that is either not working as
designed, or it stopped working at some stage.

FWIW I have set up an Azure Pipeline to keep Git for Windows' `main`
branch covered by Coverity:

https://dev.azure.com/git-for-windows/git/_build?definitionId=35

It essentially calls into this scripted code:
https://github.com/git-for-windows/build-extra/blob/4676f286a1ec830a5038b32400808a353dc6c48d/please.sh#L1820-L1915
Do you have any objection to adding something like the Action I showed
eariler? It would do nothing in git-for-windows/git unless you set up
the right environment, so there shouldn't be any downside.

I admit I was not really planning to try to suppress the false positives
as you've done here; my plan was to just keep an eye on the "new"
entries (having already gone through the existing ones years ago).

-Peff
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help