crates/stdlib/src/csv.rs (3)

1004-1024: ⚠️ Potential issue | 🟠 Major | 🏗️ Heavy lift

This skipinitialspace pre-pass rewrites valid CSV data.

split(|x| x == delimiter) is not quote-aware, so rows like a," b , c",d get mutated before csv_core parses them. It also trims trailing spaces, which skipinitialspace should not do. This needs to be handled during parsing rather than by rewriting the raw line first.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@crates/stdlib/src/csv.rs` around lines 1004 - 1024, The current pre-pass
using split + trim_spaces (functions/variables: trim_spaces, skipinitialspace,
delimiter, input) rewrites quoted fields and trims trailing spaces incorrectly
before csv_core parsing; remove this pre-pass and instead implement
skipinitialspace behavior inside the CSV parser/tokenizer (the csv_core parsing
logic) so you only ignore a single optional space immediately after a delimiter
when not inside a quoted field, without trimming trailing spaces or touching
characters inside quotes; adjust the parser's state machine to, on seeing a
delimiter and then a space, skip that single space only if not in a quoted
context, and delete the trim_spaces/split usage and the String::from_utf8
roundtrip so raw input is preserved for csv_core.

---

866-927: ⚠️ Potential issue | 🔴 Critical | ⚡ Quick win

Remove the reachable panic in writer quotechar handling.

from_args can produce quotechar = Some(None), and this branch turns that into todo!(). That makes csv.writer(..., quotechar=None) crash instead of raising a Python exception. This builder also only applies self.quoting, so a dialect's own quoting value is lost unless it came from kwargs. Resolve the effective dialect once and validate quotechar=None before building. As per coding guidelines, "When branches differ only in a value but share common logic, extract the differing value first, then call the common logic once to avoid duplicate code".

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@crates/stdlib/src/csv.rs` around lines 866 - 927, In to_writer, avoid the
reachable panic from the todo!() when self.quotechar == Some(None) (produced by
from_args) and ensure a dialect's quoting is preserved: first resolve an
effective dialect/options by merging DialectItem::Str (lookup in GLOBAL_HASHMAP)
or DialectItem::Obj with the explicit self.* overrides (delimiter, quotechar,
doublequote, lineterminator, escapechar, quoting), validate that an effective
quotechar of None is turned into a Python-level error (return a PyErr) instead
of calling todo!(), and then build the csv_core::Writer once using those
resolved values (extract the differing quotechar/quoting values first, then call
builder.delimiter(...).quote(...).double_quote(...).terminator(...).escape(...).quote_style(...)
as appropriate) so the dialect quoting is preserved when self.quoting is not
provided; refer to the to_writer function, DialectItem::Str/Obj, GLOBAL_HASHMAP,
and the fields quotechar and quoting when making these changes.

---

793-863: ⚠️ Potential issue | 🟠 Major | ⚡ Quick win

Apply the selected dialect's quoting and lineterminator here.

When self.quoting/self.lineterminator are unset, this block still forces reader settings from the override fields instead of the selected dialect. A registered/object dialect with QUOTE_NONE or a custom terminator therefore reads with the wrong configuration. Resolve one effective PyDialect first, then layer explicit overrides on top. As per coding guidelines, "When branches differ only in a value but share common logic, extract the differing value first, then call the common logic once to avoid duplicate code".

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@crates/stdlib/src/csv.rs` around lines 793 - 863, The to_reader function
currently applies reader settings per-branch and then layers overrides in a way
that ignores the selected dialect's quoting and lineterminator when those fields
are unset; fix it by first resolving one effective PyDialect (handle
DialectItem::Str via GLOBAL_HASHMAP lookup, DialectItem::Obj, and the default
"excel") to extract base values for delimiter, doublequote, quotechar, quoting,
and lineterminator, then create a single csv_core::ReaderBuilder from those
resolved base values and finally apply explicit overrides from self
(self.delimiter, self.quotechar, self.quoting, self.lineterminator,
self.doublequote, self.escapechar) only when they are Some, removing duplicated
builder logic and the redundant reader.terminator(...) at the end so the final
reader reflects the selected dialect unless explicitly overridden.