NL File format: "Method Error: convert(Expr, Float64)" when trying to parse V section of NL file

[GiovanniKarra]

When trying to parse an NL file with the latest commit of MOI, it throws a method error at this line, while trying to convert a Float64 to an Expr.
What is surprising to me is that when this parsing function wasn't implemented yet, the file was loaded without issue.

Here is a list of files where this occurs:

• deconvc.nl
• deconvu.nl
• deconvb.nl

Corresponding AMPL mod files if you find it useful:

• deconvc.mod
• deconvu.mod
• deconvb.mod

[odow]

Thanks, I'll take a look at this today.

[odow]

Fixed in #2980

What is surprising to me is that when this parsing function wasn't implemented yet, the file was loaded without issue.

Not sure what you mean by this? With the prior release of MOI, I get:

julia> using JuMP

julia> model = read_from_file("/Users/odow/Downloads/deconvc.nl")
ERROR: Unable to parse NL file : we don't support common exprs
Stacktrace:
  [1] error(s::String)
    @ Base ./error.jl:44
  [2] _parse_header(io::IOStream, model::MathOptInterface.FileFormats.NL._CacheModel)
    @ MathOptInterface.FileFormats.NL ~/.julia/packages/MathOptInterface/Q3V1z/src/FileFormats/NL/read.jl:400
  [3] read!(io::IOStream, model::MathOptInterface.FileFormats.NL.Model)
    @ MathOptInterface.FileFormats.NL ~/.julia/packages/MathOptInterface/Q3V1z/src/FileFormats/NL/read.jl:41

[GiovanniKarra]

What is surprising to me is that when this parsing function wasn't implemented yet, the file was loaded without issue.

Not sure what you mean by this? With the prior release of MOI, I get:

julia> using JuMP

julia> model = read_from_file("/Users/odow/Downloads/deconvc.nl")
ERROR: Unable to parse NL file : we don't support common exprs
Stacktrace:
[1] error(s::String)
@ Base ./error.jl:44
[2] _parse_header(io::IOStream, model::MathOptInterface.FileFormats.NL._CacheModel)
@ MathOptInterface.FileFormats.NL ~/.julia/packages/MathOptInterface/Q3V1z/src/FileFormats/NL/read.jl:400
[3] read!(io::IOStream, model::MathOptInterface.FileFormats.NL.Model)
@ MathOptInterface.FileFormats.NL ~/.julia/packages/MathOptInterface/Q3V1z/src/FileFormats/NL/read.jl:41

I was using the master branch version 1.49.0 of MOI, because I needed it for the support for common subexpressions.

(benchmarks) pkg> st MathOptInterface
Status `~/Mémoire/benchmarks/Project.toml`
  [b8f27783] MathOptInterface v1.49.0 `https://github.com/jump-dev/MathOptInterface.jl.git#master`

julia> read_from_file("CUTE-POLY/deconvc.nl")
A JuMP Model
├ solver: none
├ objective_sense: MIN_SENSE
│ └ objective_function_type: AffExpr
├ num_variables: 51
├ num_constraints: 52
│ ├ VariableRef in MOI.GreaterThan{Float64}: 51
│ └ Nonlinear: 1
└ Names registered in the model: none

But when using v1.50.0, I got the error

(CUTE) pkg> st MathOptInterface
Project CUTE v0.1.0
Status `~/CodeProjects/CUTE.jl/Project.toml`
  [b8f27783] MathOptInterface v1.50.0 `https://github.com/jump-dev/MathOptInterface.jl.git#master`

julia> CUTE.get_model("deconvc")  # calls read_from_file behind the scenes
ERROR: This model cannot be loaded by the current version of the NL file parser
MethodError(convert, (Expr, 0.0), 0x0000000000009779)
Stacktrace:
 [1] error(s::String)
   @ Base ./error.jl:44
 [2] get_model(name::String)
   @ CUTE ~/CodeProjects/CUTE.jl/src/CUTE.jl:87
 [3] top-level scope
   @ REPL[3]:1

caused by: MethodError: Cannot `convert` an object of type Float64 to an object of type Expr
The function `convert` exists, but no method is defined for this combination of argument types.

[odow]

I was using the master branch version 1.49.0 of MOI, because I needed it for the support for common subexpressions.

I can't explain this. There are only two relevant commits:

https://github.com/jump-dev/MathOptInterface.jl/commits/master/src/FileFormats/NL

Either your master was before #2938, in which case you should have received the we don't support common exprs error. Or it was after, in which case you should have received the convert error.

In no case with any commit between v1.49.0 and v1.50.0 can this have worked without erroring.

read_from_file("CUTE-POLY/deconvc.nl")

Are you sure that this file was the original and you didn't modify it somehow?

Regardless, it now works on the latest master. I'll get a new release out shortly.

[GiovanniKarra]

Are you sure that this file was the original and you didn't modify it somehow?

It is converted from a .mod file using ampl, but it is a possibility that I changed the conversion parameters at some point, I'm not sure.

Anyway, thank you very much for fixing it!