[sspickle]

Fixes two regressions from #223 (#170). Before #223, the preprocessor's "-" → "-1*" rewrite hid two RapydScript bugs, and removing it exposed both. glowscript.org has been rolled back to the version before #223 (20261101t092705) until this is merged.

1. -(x+1)**2 came out positive

RapydScript's parser gave unary operators higher precedence than **, so -x**2 parsed as (-x)**2. The output stage then turned that back into -Math.pow(x, 2), but only when the unary had no parentheses. Everything else came out with the wrong sign:

	Python	with #223	this PR
-(x+1)**2 (x = 5)	-36	36	-36
-(x)**2	-25	25	-25
-(2)**2	-4	4	-4
- -x**2	25	-25	25
exp(-(x-3)**2/8)	0.61	1.65	0.61

Fix (61d0edf): maybe_unary now parses Python's primary ** factor itself. After an atom, ** takes the next unary expression as its right operand. That makes ** bind tighter than a unary operator on its left, and makes it group right to left. That also fixes two items from #224: 2**3**2 (was 64, now 512) and 2**-1**2 (was 0.25, now 0.5). Reported upstream as kovidgoyal/rapydscript-ng#195; Kovid fixed it the same way in kovidgoyal/rapydscript-ng@50aa052 (Nov 3).

2. - -x compiled to --x

RapydScript printed a unary operator directly against its operand, so - -x became --x. In JavaScript that's a pre-decrement: it gives the wrong value and changes x. - -3 was a syntax error, so the program didn't compile at all.

Fix (b185ec0): print_unary_prefix now adds a space when a sign operator follows another sign operator. Upstream fixed this the same day as kovidgoyal/rapydscript-ng#194 (ec30155).

Testing

• tests/test_compiler.py compiles 31 expressions with the bundled compiler under node and compares each result with CPython: 10 repeated-sign cases and 21 power/unary cases. Before the fixes, 4 and 9 of them fail. After, all pass. It skips if node isn't installed, and python -m pytest --noconftest tests/test_compiler.py runs it without the Flask app.

• Through GlowScriptOffline:

  Check	Before fix(#170): stop rewriting "-" as "-1*" before compiling #223	fix(#170): stop rewriting "-" as "-1*" before compiling #223	This PR
  122-row arithmetic table, rows differing from CPython	34	15	13
  Power table (17 rows)	all match	8 wrong	all match
  Improper handling of negative signs with exponents and modulo #170 39-case program			0 failures
  23 bundled demos			all run

  The 13 rows still differing are the known %, round-half, bool and e items from Arithmetic that differs from Python: % with negatives, ** grouping, round halves, bool as int, e #224.

• **kwargs in calls and definitions, comprehensions, and conditionals containing ** all behave as before.

• The same compiler is on Port glowscript compiler fixes: randint with negatives, 3/-3, and - -x rsWVPRunner#5, and the trinket trials will run it as rsWVPRunner v3.2.4-rc3.

🤖 Generated with Claude Code

[sspickle]

On hold. #223 (#170) has a second, larger regression, and this fix doesn't cover it.

Unary minus in front of a parenthesized base of ** now has the wrong sign:

Python	Python gives	with #223
-(x+1)**2 (x = 5)	-36	36
-(x)**2	-25	25
-(2)**2	-4	4
- -x**2	25	-25
exp(-(x-3)**2/8)	0.61	1.65

Cause: RapydScript's parser binds a unary operator tighter than **. The output stage then repairs only an unparenthesized -x**2. The old "-" → "-1*" rewrite was hiding this. Every build before #223 gets all of these right.

glowscript.org has been rolled back to the version before #223 (20261101t092705). A parser fix, so that ** binds tighter than a unary minus on its left and groups right to left as in Python, will follow, with tests, before this is merged.

🤖 Generated with Claude Code

[sspickle]

Ready for review again. This now includes the second fix: ** binds tighter than a unary operator on its left and groups right to left, as in Python, so -(x+1)**2 is -36 again and 2**3**2 is 512. Upstream RapydScript fixed it the same way in kovidgoyal/rapydscript-ng@50aa052 (issue kovidgoyal/rapydscript-ng#195).

Tested with rsWVPRunner v3.2.4-rc3 (RScompiler sha256 ac91f82f…) on both trinket trial deploys:

• arithmetic spec: 123 expressions match CPython; the known differences (Arithmetic that differs from Python: % with negatives, ** grouping, round halves, bool as int, e #224) are unchanged
• full deploy suites: 95 passed on Cloud Run/Firestore and 79 passed on Mongo, no failures
• unit suites on Mongo, Garage and Firestore: no failures

🤖 Generated with Claude Code