Verifier failure

[mpickering]

The following entries file fails the verifier.

https://gist.githubusercontent.com/mpickering/70fbf67139616c4b07b1612d2346d9b9/raw/159805e72dfd2d86897ff7f6a4cd6cdfb3c40baa/Data.Distributive.Genericdistributive-0.5.3-8q2IgDtEwLpIcbv8aEzgqI

> cat /nix/store/rb2a9vkhnkjc6jf0fgfsv43dbnm0irbf-ghc-8.6.1-with-packages/cache/Data.Distributive.Genericdistributive-0.5.3-8q2IgDtEwLpIcbv8aEzgqI | /nix/store/r144r6amhd1hnwd0mvnfhqkq0q7s7p7r-kythe-0.0.28/bin/verifier -use_file_nodes -alsologtostderr -ignore-dups
Fact has invalid form:
  App(fact, ("", "", "", "", ""))
Segmentation fault

You can reproduce this if you wish using nix by

cachix use mpickering
nix-shell https://github.com/mpickering/haskell-indexer-nix/archive/38157c8c5153c561a91e88340fa297d0c573e823.tar.gz

[robinp]

Odd. The verifier is used to verify the analyser itself, by corresponding the entries that were output with special source-code annotations that prescribe what entries should be output in a textual format (see ).

Here you run it on an entry-set whose file likely doesn't have the annotations (since it is not a source prepared for this purpose). But you are right that an empty set of annotations shouldn't cause failure. Maybe the indexer outputs some invalid entry. Let me check.

[robinp]

Really odd.

$ cat test.entries | 
      /opt/kythe-v0.0.29/tools/entrystream --write_format=json | 
      /opt/kythe-v0.0.29/tools/entrystream --read_format=json > out.ents

$ cmp -b test.entries out.ents                                                                                                                    
cmp: EOF on out.ents after byte 354881, in line 3279                                   

$ hexdump -C test.entries | tail -n3
00056a30  79 74 68 65 2f 6c 6f 63  2f 65 6e 64 2a 03 37 36  |ythe/loc/end*.76|
00056a40  35 ab 01                                          |5..|

$ hexdump -C out.ents | tail -n3
00056a30  79 74 68 65 2f 6c 6f 63  2f 65 6e 64 2a 03 37 36  |ythe/loc/end*.76|
00056a40  35                                                |5|

It seems there's some garbage at the end of the entries file, which roundtripping through json silently removes.