Add lastError function

[d-torrance]

This returns a sequence containing the location and message string for the last error (at least most of the time, e.g., not for syntax errors).

The eventual goal is to use this in the MPI package (see #2129) for handling errors gracefully, e.g., doing something like try x else lastError().

[mahrud]

Are you planning on replacing the manual tests above with automatic versions?

[d-torrance]

Good idea!

[pzinn]

I did something similar on my dev branch. I'll compare the code.

[pzinn]

not directly related, but do you know what these two lines are for:

M2/M2/Macaulay2/d/expr.d

Lines 307 to 308 in 3fc8759

	export errorMessageClass := newtypeof(basicListClass);
	export missingMethodClass := newtypeof(errorMessageClass);

[d-torrance]

No idea! They don't appeared to be used anywhere.

[pzinn]

here are AFAICT the main differences I find between your implementation and mine:

• I update the last error message when it's printed, not when it's built.
• I define a new global variable lastError (well, I didn't call it that...) rather than a function lastError(). On that last point, can you comment on why this is better?

[d-torrance]

• I update the last error message when it's printed, not when it's built.

One of the big goals was to find out if there was an error without actually raising one, e.g., try foo() else lastError() would tell us what happened if foo() raised and error without actually raising an error.

That being said, @MichaelABurr has some code updating it during printErrorMessage. See d-torrance#13. Is that what you were doing, too?

• I define a new global variable lastError (well, I didn't call it that...) rather than a function lastError(). On that last point, can you comment on why this is better?

That was our original idea, but we didn't realize it was possible. Most of the global variables like this (e.g., printWidth) seem to be set at top-level and accessed in the interpreter. Whereas this would be set in the interpreter and accessed at top-level. We couldn't figure out an obvious way to do that, hence the function.

Is your code live on Macaulay2Web? Or just in a branch somewhere?

[pzinn]

good point. maybe I'm wrong about updating it when it's printed, as you said, the main goal is to test the error message with try. Yes it is live on M2Web, but it's not working any longer, I made this at the same time as my (failed) error upgrade exactly a year ago, and since then changes to the official development branch broke it; try viewHelp "try", the example there doesn't work any more.

[pzinn]

the relevant commit seems to be 74e022a
yeah it's clumsy, it updates the error message when it's printed or when try is called (which I'm assuming is the part that got overwritten by a subsequent PR).

[pzinn]

I still think having a global variable is marginally simpler than a function.
Ah, another thing is I also defined the location of the last error. Not sure if it belongs in your PR.
The goal for me was to transfer all the error printing to the m2 level.

[mahrud]

It might be nice to have something another variable to remember the value of current after an error has occurred.

[d-torrance]

I've rewritten this a bit so that lastError is a variable, like Paul's version, that we can clear by assigning null:

i1 : 1/0
stdio:1:1:(3): error: division by zero

i2 : lastError

o2 = (stdio:1:0-1:3, division by zero)

o2 : Sequence

i3 : lastError = null