Added missing 3rd (of 4) arg to jsnums.toRepeatingDecimal() calls #556

[ds26gte]

These occur in ide.js and output-ui.js.

[blerner]

The code change looks reasonable. But do we have a test to make sure this is correct? The original issue definitely had something that triggered this bug, so we should be able to extract a regression test for this...

[schanzer]

@blerner @ds26gte here's the file that exposes the bug (evaluate Cell-TABLE after running). Here's the offending spreadsheet.

[jpolitz]

This seems like much more than a missing errback issue: somehow the parsing of this sheet is leading to a rational with a 0 denominator. The string renderers show it, the toRepeatingDecimal call in the DOM renderer errors:

https://code.pyret.org/editor#share=1a1Pd-fdFlF6rj3QfkDP-St8LNbhzdV8r&v=fee2ecd

The data is:

Molecules	Species	Speed
Na+	ions	5.00E-19

So somehow in the sheets reader we are getting 1298074214633707/0 as a value. In the debugger that value looks like this:

[jpolitz]

Sheet URL: https://docs.google.com/spreadsheets/d/1MDCFhQl2tw-FdALuBS8vXC3Kom11aMocgD1OPVS256c/edit?gid=0#gid=0

[jpolitz]

Note that the effect of the code in this PR is just to raise a nicely rendered DomainError, which is good, but should not happen on this input.

[blerner]

This sounds like something that's easily reproducible as a JS regression test, then? I.e. take the string 5E-19 (or 5.00E-19) and parse it and make sure we don't get a broken number?

[jpolitz]

The constant in a Pyret program is fine. It's something in the sheets reading/parsing.

[blerner]

But I think the field content is actually 5E-19, without the .00 -- I suspect the cell formatting is showing two decimal places. And in CPO,

According to Pyret, the value inside that some has a to-repr of 5e-19, a to-string of 5e-19, and yet renders as nothing at all in the output... So something is very strange with that value.

[jpolitz]

What browser are you on?

Brave Version 1.76.82 Chromium: 134.0.6998.178 (Official Build) (arm64)

[blerner]

Hmmm, I may have had stale state in my browser window somehow; refreshing made the problem go away, so /shrug?

[jpolitz]

OK, the culprit is fromFixnum in js-numbers, which is somehow wrong (IDK how yet) in the logic that was added for handling scientific notation literals:

brownplt/pyret-lang@8047792#diff-34cd626ab11424687bfd071b24342ee9d60c879050d69239ca8cf86df95e2f39R7-R77

fromFixnum is relatively hard to reach from Pyret (most things use makeNumberFromString which uses jsnums.fromString). However, json uses it, so a simple command-line repro is:

→ cat json-5e.arr
import json as J

print(J.read-json("5E-19"))
→ make json-5e.jarr
→ node json-5e.jarr
j-num(1298074214633707/0)

[jpolitz]

@ds26gte if you're still on this: a useful way to proceed would be to figure out how to write tests for fromFixnum (which likely was never tested very thoroughly), perhaps using json, and then make sure it works for many cases with E literals.