[test/spec] Split history builtin tests from builtins2 tests

[PossiblyAShrub]

As per #1587 (comment).

---

While working on this I realized we were missing the case where HISTFILE does not point to a file (even if it's a valid string). This should return an error code if a user tries to then history -r.

[andychu]

hm actually ErrExit is intended for use with set -o errexit

I think simply return 1 will do what you want here?

Also I would just say if not path_stat.exists() instead of the extra var

---

Is this intended to be friendlier than bash? I think bash just ignores the error?

I tried bash -i -c 'HISTFILE=a; history -r

[PossiblyAShrub]

Ok. I still want to provide some information about the failure. I see that we use a self.errfmt.Print_ pattern in other builtins. Would that be appropriate here?

Is this intended to be friendlier than bash? I think bash just ignores the error?

I tried bash -i -c 'HISTFILE=a; history -r

I do get an error code on my machine (no message though)

$ HISTFILE=_tmp/dne
$ history -r
$ echo $?
1

[andychu]

Ah OK , makes sense.

Yes errfmt.Print_ is consistent

---

I changed the error message to use %r because that will quote the filename properly (e.g. in the rare case that the filename contains a single quote)

In the C++ runtime we have a format string implementation that supports %s %r %d and that's it basically

• %s and %r must be strings (unfortunately you get a runtime error if not, not compile time
• %d must be an integer

[andychu]

Merged, thanks for doing this!