Improve parse_eval_all line number reporting

[ihnorton]

Wraps a LoadError thrown by jl_parse_eval_all with the original top-level line/file from the parser.

Ref: jump-dev/JuMP.jl#7 where we have a macro expansion that will throw UndefVarError:

using JuMP
m = Model()
@setObjective(m, Max, z)

before:

ERROR: LoadError: UndefVarError: z not defined
 in include at ./boot.jl:260
 in include_from_node1 at ./loading.jl:304
while loading /cmn/julia/dummy.jl, in expression starting on line 584

after:

julia> include("dummy.jl")
ERROR: LoadError: LoadError: UndefVarError: z not defined
 in include at ./boot.jl:261
 in include_from_node1 at ./loading.jl:304
while loading /home/isaiah/.julia/v0.5/JuMP/src/macros.jl, in expression starting on line 584
while loading /cmn/julia/dummy.jl, in expression starting on line 3

[ihnorton]

(I realize that the repetitive while loading and LoadError: LoadError messages are not ideal. Maybe we should add a wrapper error type that displays like a backtrace entry?)

[ihnorton]

Rebased and passes tests now.

[jakebolewski]

This isn't quite correct, it doesn't handle recursive include's correctly so you end up with a bunch of -1 line numbers being reported.

One improvement is to move, line 578 of this diff needs to be move right after jl_parse_next(), that will get the last load error to report the correct line number.

Ex:

LoadError(at "sysimg.jl" line -1: LoadError(at "sysimg.jl" line 241: LoadError(at "linalg.jl" line -1: LoadError(at "linalg.jl" line 225: LoadError(at "linalg/cholesky.jl" line -1: LoadError(at "linalg/cholesky.jl" line 26: UndefVarError(var=:Typer)))))))

I also think the load error's need to be improved. Tracing through where the file is included from is pretty useful, but the output needs to be printed in a better way.

[jakebolewski]

Sorry for the noise on master, but this is more obvious during bootstrap.

[ihnorton]

it doesn't handle recursive include's correctly so you end up with a bunch of -1 line numbers being reported.

I didn't see your issue or this because it works fine in a debug build (note that top_lineno is a local variable).
I added the line a = Typer= at linalg/cholesky.jl:25, and get the following with make debug:

linalg.jl
error during bootstrap:
LoadError(at "sysimg.jl" line 5: LoadError(at "sysimg.jl" line 241: LoadError(at "linalg.jl" line 3: LoadError(at "linalg.jl" line 225: LoadError(at "linalg/cholesky.jl" line 25: UndefVarError(var=:Typer))))))

whereas in a release build:

linalg.jl
error during bootstrap:
LoadError(at "sysimg.jl" line -1: LoadError(at "sysimg.jl" line 241: LoadError(at "linalg.jl" line -1: LoadError(at "linalg.jl" line 225: LoadError(at "linalg/cholesky.jl" line -1: LoadError(at "linalg/cholesky.jl" line 25: UndefVarError(var=:Typer)))))))

Maybe top_lineno is not properly restored by the catch handler under some conditions?
(when I step through the release build in gdb, it is set correctly when I return from jl_parse_next, but then reverts to -1 when the expression is evaluated / error thrown)

[ihnorton]

I also think the load error's need to be improved. Tracing through where the file is included from is pretty useful, but the output needs to be printed in a better way.

Isn't this mostly an issue during bootstrap? The output is somewhat better in the REPL (see example in the OP).