doc: warn about reassigning _ in REPL

[thefourtheye]

When users assign a value to _ in REPL, it is prone to unexpected
results or messing up with REPL's internals. For example,
#3704. This patch issues a warning
about the same.

[SamuelMarks]

@thefourtheye how about going one step further, and throwing an error or printing a warning when _ assignment is attempted?

Looks straightforward enough to do, just follow this pattern: repl.js#L384

[thefourtheye]

@SamuelMarks The assignment to _ expression will be a string. You meant to say, parse the string and check if it is assigning something to _?

[Fishrock123]

What is _ used for that this happens? (We could probably make it something else?)

[thefourtheye]

@Fishrock123 Whenever you evaluate something in the REPL, the last evaluated expression's result is stored in _. For example,

> 1 + 1
2
> _ + 3
5
> _
5

If we assign something else to _, then it will mess up the further evaluations in REPL. For example,

> 1 + 1
2
> _
2
> _ = -2
-2
> _ + 2
0

Even worse, if we make _ a const, REPL will totally break, as it will try to assign the last evaluated expression's result in _. For example,

> const _ = 1;
TypeError: Cannot assign to read only property '_' of #<Object>
    at finish (repl.js:457:24)
    at REPLServer.defaultEval (repl.js:269:5)
    at bound (domain.js:280:14)
    at REPLServer.runBound [as eval] (domain.js:293:12)
    at REPLServer.<anonymous> (repl.js:412:12)
    ...
> ;
TypeError: Cannot assign to read only property '_' of #<Object>
    at finish (repl.js:457:24)
    at REPLServer.defaultEval (repl.js:269:5)
    at bound (domain.js:280:14)
    at REPLServer.runBound [as eval] (domain.js:293:12)
    at REPLServer.<anonymous> (repl.js:412:12)
    ...

---

_ is the commonly used name for this purpose I guess. For example, Python's REPL also uses _ only

Python 2.7.6 (default, Jun 22 2015, 17:58:13) 
[GCC 4.8.2] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> 1 + 1
2
>>> _
2
>>> _ + 3
5
>>> _
5
>>> _ = -1
>>> _
-1

[Fishrock123]

@thefourtheye Perhaps we can catch assigning to it while parsing and warn?

[thefourtheye]

@Fishrock123 I think we can allow users to assign to it (I couldn't think of a case where assigning to it could be a very big trouble), but defining it as const as the first line, would totally break our REPL. What we can possibly do is, if user defines _ explicitly as const in REPL, we can stop storing the last value in _. Will that be okay?

[cjihrig]

I think we should block the creation of a const _ (and show an appropriate message). That seems to be the only case that disrupts the REPL functionality.

[jasnell]

If we pre-init _ to undefined then calling const _ = ought to result in an Identifier '_' has already been declared error.

[targos]

👍 for @jasnell's suggestion

[thefourtheye]

PR coming soon with @jasnell's awesome suggestion :-)

[jasnell]

Perhaps simply: *NOTE*: Explicitly assigning a value to_in the REPL can produce unexpected results.

[thefourtheye]

@jasnell I updated the text as you suggested. Thanks :-) PTAL.

[jasnell]

LGTM

[cjihrig]

LGTM, but this should probably just be rolled into the same commit as #3737

[thefourtheye]

@cjihrig You are correct. Moving these changes to #3737 and closing this.

[MylesBorins]

removing from lts-watch in lieu of #3737