adjust parsing to raise nicer exceptions

[snoyer]

The current parsing works great on well formed path but could be improved when it comes to handling invalid cases:

1. the "Unallowed implicit command" error message mentions an internal token index
2. invalid tokens are silently dropped
3. the parser can run out of tokens and let an IndexError: pop from empty list through

Path('1,2')
ValueError: Unallowed implicit command in 1,2, position -1

Path('M 100 100 L 200 200 Z 100 200')
ValueError: Unallowed implicit command in M 100 100 L 200 200 Z 100 200, position 7

Path('M 0 1 L 0 0 L 0 0z1')
ValueError: Unallowed implicit command in M 0 1 L 0 0 L 0 0z1, position 8

Path('M 1 2 3')
IndexError: pop from empty list

Path('M 0 1 N 2 3')
Path(Line(start=1j, end=(2+3j)))

Path('M 0 1 C 1 2 3 4\n      5 foo')
IndexError: pop from empty list

This PR catches cases 2. and 3. and raises a standard SyntaxError highlighting the error location in the d string in all 3 cases:

Path('1,2')
  File "<svg-d-string>", line 1
    1,2
    ^
SyntaxError: missing command

Path('M 100 100 L 200 200 Z 100 200')
  File "<svg-d-string>", line 1
    M 100 100 L 200 200 Z 100 200
                          ^^^
SyntaxError: missing command

Path('M 0 1 L 0 0 L 0 0z1')
  File "<svg-d-string>", line 1
    M 0 1 L 0 0 L 0 0z1
                      ^
SyntaxError: missing command

Path('M 1 2 3')
  File "<svg-d-string>", line 1
    M 1 2 3
          ^
SyntaxError: not enough arguments

Path('M 0 1 N 2 3')
  File "<svg-d-string>", line 1
    M 0 1 N 2 3
         ^^^
SyntaxError: invalid token ' N '

Path('M 0 1 C 1 2 3 4\n      5 foo')
  File "<svg-d-string>", line 2
    5 foo
     ^^^^
SyntaxError: invalid token ' foo'

[mathandy]

Closing due to lack of activity.

[snoyer]

Closing due to lack of activity.

Should I rebase for some review activity or is this PR not of interest?

[mathandy]

Hey @snoyer thanks for the PR and for the followup after I closed it.

I like the feature/direction, looking at it, I think it's going to require some work. If you want to rebase and fix up the below issues, I'll take another look and I think we can merge something in along these lines.

I had claude write up a review, it summarized a lot of my quick impressions pretty well. If you have any questions, I can add more explanation.

# Review: adjust parsing to raise nicer exceptions

*Reviewed on the PR's own base (all 10 `test_parsing` tests pass there). Note the base now predates two merged PRs (#245, #246) that touch the same code — see the conflict notes below.*

## Overview

This PR rewrites `Path._parse_path`/`_tokenize_path` error handling so malformed `d`-strings raise a `SyntaxError` that pinpoints the offending token with line/column info and a caret, instead of the current mix of a confusing `ValueError` (with an internal token index), silently dropped garbage tokens, and bare `IndexError: pop from empty list`. Mechanically: the tokenizer emits `(token, lineno, offset)` tuples including *invalid* tokens (nothing silently dropped), all `float(elements.pop())` calls become a `pop_float()` helper that converts failures into located errors, and a `_syntax_error` helper builds the `SyntaxError` with a py<3.10 fallback. Three targeted error tests replace the single old one.

The goal is genuinely good — error messages with carets into the `d`-string are a real usability win, and no longer silently ignoring `M 0 1 N 2 3` fixes silent-corruption behavior. But there is one design problem and a few bugs to resolve.

## Issues

### Blocking / design

- **Breaking exception-type change.** `SyntaxError` is not a subclass of `ValueError`, so any caller doing `except ValueError` around `parse_path` breaks (verified on the branch). It also disagrees with master, where #246 now raises `ValueError` for the truncated case. A `class PathSyntaxError(SyntaxError, ValueError)` would keep the caret rendering *and* backward compatibility; as written this needs at least a changelog entry.
- **Conflicts with merged #245 (packed arc flags) — the rebase is not mechanical.** Master's tokenizer now special-cases arc arguments so `a1.6,1.6 0 01-1.1 2` parses (`_tokenize_arc_args`, `ARC_FLAG_RE`). This PR replaces the tokenizer wholesale with a single `TOKEN_RE`, which lexes `01` as one float — on this branch that input fails with "not enough arguments". A rebase must port flag-aware arc tokenizing into the new position-tracking tokenizer; #246's `COMMAND_NUM_ARGS` check then becomes redundant and can be removed in favor of `pop_float`.

### Bugs

- **Noisy chained tracebacks.** `raise self._syntax_error(...)` inside `except ValueError:`/`except IndexError:` produces "During handling of the above exception, another exception occurred" with the internal `float()`/`pop()` traceback shown first (verified). Needs `raise ... from None`.
- **`end_lineno` off-by-one in `_syntax_error`.** The 6-tuple is `(filename, lineno+1, offset+1, line, lineno, end_offset+1)` — the 5th element is `end_lineno`, which must be 1-based like the 2nd; passing the 0-based `lineno` makes `end_lineno < lineno`. Should be `lineno + 1`.
- **Empty-last-line edge case.** In the `IndexError` branch, `end = len(line) - 1` gives `-1` when the last line is empty (e.g. `parse_path('M 1\n\n')`), producing offset 0 — and the caret points at the wrong line anyway. Better: walk back to the last non-empty line, or reuse the last-popped token's position.

### Minor / style

- Error spans include raw separators — `invalid token ' N '` with the caret span covering the spaces. Trimming to the stripped token would be cleaner.
- `_tokenize_path(*pathdef_lines)` varargs is odd for a private helper; just take the list.
- `def _syntax_error(cls, msg, lineno, line, offset,end_offset)` — missing space (PEP 8); the `except IndexError` fallback for py<3.10 deserves a comment explaining *why* the 6-tuple raises there.
- Tests use bare `assert` inside `unittest` classes; the file's convention is `self.assert*` (e.g. `self.assertIn("missing command", e.exception.msg)`). Also missing: a test that a valid multi-line path still parses (offset bookkeeping), and — post-rebase — the packed-arc-flag cases from #245's tests must keep passing.

## Verdict

Worth reviving — located syntax errors are strictly better than what master has, and #246 only addressed one of the three failure modes this covers. But it can't merge as-is: it needs a rebase that reworks the tokenizer around #245's arc-flag handling, a decision on exception type (recommend a `SyntaxError`+`ValueError` hybrid for compatibility), and the bug fixes above. The PR's tests and design give a clear spec for that work.

[snoyer]

If we're going to be engaging in this kind of proxying you might as well have claude implement the fix directly.