Fix raw_values mutation that broke repeated raw= searches

[roed-math]

The bug

search() and lucky() declare a mutable default raw_values=[], and _build_query aliases that list before appending the limit/offset:

qstr, values = SQL(raw), raw_values   # values IS the caller's (default) list
...
if limit is not None:
    values.append(limit)              # mutates it
    if offset != 0:
        values.append(offset)

So the first raw= search in a process works, but every subsequent one inherits the accumulated limit/offset from the shared default list. The generated SQL then has fewer %s placeholders than values, and the query dies with:

TypeError: not all arguments converted during string formatting

A caller that passes its own raw_values list is also affected — the list is mutated as a side effect of the search.

Reproduce

from lmfdb import db
nf = db.nf_fields
list(nf.search({}, projection=["label"], limit=3, raw="degree = 2"))  # ok
list(nf.search({}, projection=["label"], limit=3, raw="degree = 2"))  # TypeError

The fix

• Copy raw_values in _build_query (values = list(raw_values)) so the caller's list is never mutated.
• Change the mutable defaults in search() and lucky() from [] to None (_build_query already coalesces None to []).

After the fix, repeated raw= searches succeed and a caller-supplied raw_values list is left untouched. Added a doctest in _build_query demonstrating both.

(Found while adding a command-line raw-SQL search to the LMFDB.)

🤖 Generated with Claude Code