[ 
https://issues.apache.org/jira/browse/ARROW-16385?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17529669#comment-17529669
 ] 

Kouhei Sutou commented on ARROW-16385:
--------------------------------------

I agree with [~apitrou].

For this case, can we detect Snappy version and whether sanitizer is enabled or 
not? We'll be able to write more descriptive skip logic for this with these 
information. (We skip the test only with Snappy 1.1.8 && sanitizer is enabled.)

> [R] [CI] Clean up our snappy-sanitizer skipping behavior
> --------------------------------------------------------
>
>                 Key: ARROW-16385
>                 URL: https://issues.apache.org/jira/browse/ARROW-16385
>             Project: Apache Arrow
>          Issue Type: Bug
>          Components: Continuous Integration, R
>            Reporter: Jonathan Keane
>            Assignee: Jacob Wujciak-Jens
>            Priority: Major
>
> We have a number of locations where we skip parquet tests now that snappy is 
> built by default + we use it by default when it is built.
> One recent example of needing to do this is 
> https://github.com/apache/arrow/pull/13014
> However, skipping tests like this is a little bit of misdirection, since we 
> aren't really skipping these because | when snappy is not available like the 
> helper suggests, just using that helper to _also_ skip when we know we are in 
> a sanitizer environment.
> The ultimate answer to this, of course is to upstream the change 
> https://github.com/google/snappy/pull/148 though that's been sitting open for 
> a few months still.
> In the meantime, what if we took out these skips and instead used 
> uncompressed parquet for reading and writting in some builds? This way we 
> could make sure that snappy was not running during sanitizer tests, but still 
> have test coverage for these code paths in other runs where we don't need to 
> worry about this sanitizer error in snappy.
> https://github.com/apache/arrow/pull/13014#discussion_r859970907 proposed one 
> way to do this in this one case, but we should do it more generally for the 
> other skips that we have had to add.



--
This message was sent by Atlassian Jira
(v8.20.7#820007)

Reply via email to