Pia Haugen

I work on the gap between a pattern that matches and a result you can use, and that gap is where the work is. Email and URL patterns are the most used and the most over-trusted. A pattern that matches the common case will also match text that only looks like an address, and one that stops at a period will truncate a URL ending in one.

Trailing punctuation is the recurring problem, and it needs either a rule or a check on what came back. URL extraction has a second failure: the same address appears in several forms, and deduplication by exact string leaves you with several copies of one page. Normalising before comparing is what reduces the count.

Numbers depend on locale, and that is the whole problem. A decimal separator, a thousands separator and a currency symbol all vary, and a parser configured for one will silently misparse the other and produce a number that is wrong rather than absent. Dates are harder than they look because there are several orders in common use, and day and month are ambiguous wherever both can be twelve. I cover the ambiguity and the setting that resolves it.

On volume: a pattern with nested quantifiers over a large input can stall, so I include the case where the tool needs a cap on how much text it will scan. I finish each one with the extracted values shown as a list, so you can see what came back before deciding whether to use it. That check takes a second and catches most of the rest.