Fixed parsing of empty sparsevec in SparseVector

[OsamaAnsar]

new SparseVector('{}/3') (the text pgvector returns for an all-zero sparsevec) parses to NaN instead of an empty vector. ''.split(',') is [''], so #fromSql pushes parseInt('') - 1 and parseFloat(undefined).

Real output on current master (run against src/index.js):

new SparseVector([0, 0, 0]).toPostgres()  -> {}/3
new SparseVector('{}/3')                  -> dimensions 3, indices [ NaN ], values [ NaN ]
  .toPostgres()                           -> {NaN:NaN}/3
  .toArray()                              -> [ 0, 0, 0, NaN: NaN ]   (own keys: '0', '1', '2', 'NaN')

With this change:

new SparseVector('{}/3')                  -> dimensions 3, indices [], values []
  .toPostgres()                           -> {}/3
  .toArray()                              -> [ 0, 0, 0 ]             (own keys: '0', '1', '2')

The fix takes the text between the braces and only loops over elements if it is not empty. Non-empty input like {1:1,3:2}/4 parses as before.

Added a fromSqlEmpty test to tests/sparse-vector.test.js. It is a pure unit test and does not need a database.

What I ran:

• node --test tests/sparse-vector.test.js tests/utils.test.js: 8 pass.
• With only src/sparse-vector.js reverted to master, the new test fails (actual: [ 0, 0, 0, NaN: NaN ]) and the other 5 pass; with the fix restored all 6 pass.
• npm run build and npm run check (after prisma generate) complete without errors.
• I did not run the rest of the test suite (the ORM/driver tests) because it needs a Postgres server with pgvector, which I don't have here.

[ankane]

Hi @OsamaAnsar, thanks for the PR. Merged a version of this in the commit above.

[OsamaAnsar]

Hi @OsamaAnsar, thanks for the PR. Merged a version of this in the commit above.

Thanks!