[GitHub] [couchdb] nickva removed a comment on pull request #3488: Fix libicu inconsistency

[GitBox <gi...@apache.org>]

nickva removed a comment on pull request #3488:
URL: https://github.com/apache/couchdb/pull/3488#issuecomment-812699068


   It was interesting that we didn't hit this issue on CentOS 7 on 3.x so I was curious and traced the `less` call when running that mango test:
   
   ```
   10:9:04.429561 <0.11140.1> couch_ejson_compare:less_nif([{<<"ÿÿÿÿ">>}], [{[{<<"forename">>,<<"Eddie">>}]}])
   
   10:9:04.429783 <0.11140.1> couch_ejson_compare:less_nif/2 error badarg
   
   10:9:04.429984 <0.11140.1> couch_ejson_compare:less_erl([{<<"ÿÿÿÿ">>}], [{[{<<"forename">>,<<"Eddie">>}]}])
   
   10:9:04.430254 <0.11140.1> couch_ejson_compare:less_list([{<<"ÿÿÿÿ">>}], [{[{<<"forename">>,<<"Eddie">>}]}])
   
   10:9:04.430506 <0.11140.1> couch_ejson_compare:less_erl({<<"ÿÿÿÿ">>}, {[{<<"forename">>,<<"Eddie">>}]})
   
   10:9:04.430745 <0.11140.1> couch_ejson_compare:less_erl/2 --> 1
   
   10:9:04.430903 <0.11140.1> couch_ejson_compare:less_list/2 --> 1
   
   10:9:04.431088 <0.11140.1> couch_ejson_compare:less_erl/2 --> 1
   ```
   
   Apparently in `3.x` we compare an object `{[{<<"forename">>,<<"Eddie">>}]}` against a (highest?) marker, but that marker is not an object but a tuple with one element `{<<"ÿÿÿÿ">>}`. `less_nif/2` then notices the object is not valid ejson and throws a `badarg` error.  The the code then falls back to Erlang using `less_erl/2` and at least in this case return the correct result (highest marker is greater than (`1`) than the object).
   
   In `main` we compare against an actual object `{[{<<"ÿÿÿÿ">>, <<>>}]}` so everything gets unpacked nicely then the old unicode library at least older than 59 (and possible 53) didn't error out, but also didn't sort the `<<255,255,255,255>>` as the highest element as intended. Here is centos 7 with libicu-50.2-4.el7_7.x86_64, for example
   ```
   couch_ejson_compare:less(<<"forename">>, <<255,255,255,255>>).
   1
   > couch_ejson_compare:less(<<"/">>, <<255,255,255,255>>).
   -1
   > couch_ejson_compare:less(<<"0">>, <<255,255,255,255>>).
   1
   ```
   
   and on mac os with icu4c 59:
   ```
   > couch_ejson_compare:less(<<"forename">>, <<255,255,255,255>>).
   -1
   > couch_ejson_compare:less(<<"/">>, <<255,255,255,255>>).
   -1
   > couch_ejson_compare:less(<<"0">>, <<255,255,255,255>>).
   -1
   ```
   
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org