Unsupported: '^' from integregular when string has pattern in json schema

[ekagra-ranjan]

Describe the issue as clearly as possible:

Hi! When I use json schema in outlines that has regex pattern inside a string type, it tries to run interegular.parse_pattern(regex_string) and fails with Unsupported: '^'. It happens when regex pattern inside string has ^ or $.

Example json schema

{
  "type": "object",
  "properties": {
    "address": {
      "type": "object",
      "properties": {
        "postalCode": {
          "type": "string",
          "pattern": "^\d{5}$"
        }
      }
    }
  }
}

This seems like a limitation of interegular but Outlines supports pattern within string type so was wondering if regex pattern in json schema is really supported or not?

Also, it seems we wrap the pattern inside ^ and $ with double quotes. This would convert ^\d{5}$ to ^"\d{5}"$ . We can let the user decide if its string or numbers?

cc: @lapp0

Steps/code to reproduce the bug:

import interegular
pattern = "^\d{5}$"
new_pattern = rf'(^"{pattern[1:-1]}"$)' 
print(new_pattern)
regex_pattern = interegular.parse_pattern(new_pattern)

Expected result:

no error

Error message:

Unsupported: '^'

Outlines/Python version information:

Version information

``` (command output here) ```

Context for the issue:

No response

[ekagra-ranjan]

Hi @brandonwillard @lapp0 can someone confirm this please? Thanks!

[lapp0]

Hi @ekagra-ranjan

Interegular explicitly doesn't support ^ or $ and has tests asserting this https://github.com/MegaIng/interegular/blob/758f83721c1ac306b96b7a706b69716a67ed954b/tests/test_patterns.py#L83-L84

I'm not sure why outlines converts ^foo$ into ^"foo"$, this seems to be wrong, they should probably be stripped.

Are you able to make your pattern work by removing the illegal ^ and $?

[ekagra-ranjan]

Are you able to make your pattern work by removing the illegal ^ and $?

yes, that works.

I'm not sure why outlines converts ^foo$ into ^"foo"$, this seems to be wrong, they should probably be stripped.

this is what confuses me too. I guess this was not intended and should be been stripped as you said