rpad: value mismatch between Presto and Velox

[peterenescu]

Description

rpad produces mismatched values between Presto/Velox which propagates invalid to_base64 value in fuzzer run.

Reproduction

(Velox followed by Presto)

presto:di> select rpad(hmac_sha1(c1, to_big_endian_32((INTEGER '520535002' + (- c2)))), levenshtein_distance(reverse(varchar '{V$~ZG)uW|?Z?1W'), c3), spooky_hash_v2_64(c4)) from (values (varbinary '(D', varbinary 'XSmo6^)oy''Zhx#G{G5%1/&)9kAd%1tvqrHqclkc&AmEoXxj4?.$bb%?^''y~^EZ)t@p)kH3_()%{rg=y[3M%,e0BfY?R[)xO>', 1772733589, varchar '?VV?v4Mt28hy0^qX-wG[tc0Ejf,xn1oK~npx2)$CB5_BfD~9>uG2AY0L''bBaQch8', varbinary 'Y')) t(c0, c1, c2, c3, c4);
                      _col0                      
-------------------------------------------------
 aa 57 d9 2a da ae c6 83 33 a1 75 45 3d 9d 05 54 
 d2 e0 35 65 98 5c aa aa 46 29 5d 21 98 5c aa aa 
 46 29 5d 21 98 5c aa aa 46 29 5d 21 98 5c aa aa 
 46 29 5d 21 98 5c aa aa 46 29 5d 21 98 5c aa aa 
 46 29 5d 21 98 5c aa aa 46 29 5d 21 98 5c aa aa 
 46 29 5d 21 98 5c aa aa 46 29 5d 21             
(1 row)
presto:di> select rpad(hmac_sha1(varbinary 'XSmo6^)oy''Zhx#G{G5%1/&)9kAd%1tvqrHqclkc&AmEoXxj4?.$bb%?^''y~^EZ)t@p)kH3_()%{rg=y[3M%,e0BfY?R[)xO>', to_big_endian_32((INTEGER '520535002' + (- 1772733589)))), levenshtein_distance(reverse(varchar '{V$~ZG)uW|?Z?1W'), varchar '?VV?v4Mt28hy0^qX-wG[tc0Ejf,xn1oK~npx2)$CB5_BfD~9>uG2AY0L''bBaQch8'), spooky_hash_v2_64(varbinary 'Y'));
                      _col0                      
-------------------------------------------------
 aa 57 d9 2a da ae c6 83 33 a1 75 45 3d 9d 05 54 
 d2 e0 35 65 98 5c aa aa 46 29 5d 21 98 5c aa aa 
 46 29 5d 21 98 5c aa aa 46 29 5d 21 98 5c aa aa 
 46 29 5d 21 98 5c aa aa 46 29 5d 21             
(1 row)

Relevant logs

Some additional runs to show nested calls have rpad as the culprit of the value differences. hmac_sha1 matches (Velox followed by Presto)):

presto:di> select hmac_sha1(c1, to_big_endian_32((INTEGER '520535002' + (- c2)))) from (values (varbinary '(D', varbinary 'XSmo6^)oy''Zhx#G{G5%1/&)9kAd%1tvqrHqclkc&AmEoXxj4?.$bb%?^''y~^EZ)t@p)kH3_()%{rg=y[3M%,e0BfY?R[)xO>', 1772733589)) t(c0, c1, c2);
                      _col0                      
-------------------------------------------------
 aa 57 d9 2a da ae c6 83 33 a1 75 45 3d 9d 05 54 
 d2 e0 35 65                                     
(1 row)
presto:di> select hmac_sha1(varbinary 'XSmo6^)oy''Zhx#G{G5%1/&)9kAd%1tvqrHqclkc&AmEoXxj4?.$bb%?^''y~^EZ)t@p)kH3_()%{rg=y[3M%,e0BfY?R[)xO>', to_big_endian_32((INTEGER '520535002' + (- 1772733589))));
                      _col0                      
-------------------------------------------------
 aa 57 d9 2a da ae c6 83 33 a1 75 45 3d 9d 05 54 
 d2 e0 35 65                                     
(1 row)

to_base64 mismatches but because of mismatched input cascaed from rpad run (Velox followed by Presto):

presto:di> SELECT to_base64(hmac_sha256(c0, rpad(hmac_sha1(c1, to_big_endian_32((INTEGER '520535002' + (- c2)))), levenshtein_distance(reverse(varchar '{V$~ZG)uW|?Z?1W'), c3), spooky_hash_v2_64(c4)))) from (values (varbinary '(D', varbinary 'XSmo6^)oy''Zhx#G{G5%1/&)9kAd%1tvqrHqclkc&AmEoXxj4?.$bb%?^''y~^EZ)t@p)kH3_()%{rg=y[3M%,e0BfY?R[)xO>', 1772733589, varchar '?VV?v4Mt28hy0^qX-wG[tc0Ejf,xn1oK~npx2)$CB5_BfD~9>uG2AY0L''bBaQch8', varbinary 'Y')) t(c0, c1, c2, c3, c4);
                    _col0                     
----------------------------------------------
 4fH6sTrTZQrRf9VXR9Gx15rY/clorhuuiWWoRb3mHsQ= 
(1 row)
presto:di> SELECT to_base64(hmac_sha256(varbinary '(D', rpad(hmac_sha1(varbinary 'XSmo6^)oy''Zhx#G{G5%1/&)9kAd%1tvqrHqclkc&AmEoXxj4?.$bb%?^''y~^EZ)t@p)kH3_()%{rg=y[3M%,e0BfY?R[)xO>', to_big_endian_32((INTEGER '520535002' + (- 1772733589)))), levenshtein_distance(reverse(varchar '{V$~ZG)uW|?Z?1W'), varchar '?VV?v4Mt28hy0^qX-wG[tc0Ejf,xn1oK~npx2)$CB5_BfD~9>uG2AY0L''bBaQch8'), spooky_hash_v2_64(varbinary 'Y'))));
                    _col0                     
----------------------------------------------
 ZvCM1VjZw2OiZN19hNlVbd5PzMujGtykCr6caST+KCY= 
(1 row)

[kagamiori]

Hi @peterenescu, thank you for creating this issue. Could the example queries be simplified further, e.g., by removing the irrelevant subexpressions, or replace subexpressions with their results as literal constants? Also, there are multiple example queries, which one should I look at if I'd like to understand the difference between the rpad in Velox and that in Presto?

[peterenescu]

Just updated and moved extraneous queries to Relevant logs in case the reproducer will like the full fuzzer run queries.