kwargs are annoying in the debugger

[Keno]

Here's a transcript of what I had to do to get through a bunch of kwargs:

1|debug > n
In attach_optimizer(model) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Utilities/cachingoptimizer.jl:128
128       @assert model.state == EMPTY_OPTIMIZER
129       # We do not need to copy names because name-related operations are handled by `m.model_cache`
130       indexmap = MOI.copy_to(model.optimizer, model.model_cache, copy_names=false)
131       model.state = ATTACHED_OPTIMIZER
132       # MOI does not define the type of index_map, so we have to copy it into a

About to run: (Core.apply_type)(NamedTuple, (:copy_names,))
1|debug > nc
In attach_optimizer(model) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Utilities/cachingoptimizer.jl:128
128       @assert model.state == EMPTY_OPTIMIZER
129       # We do not need to copy names because name-related operations are handled by `m.model_cache`
130       indexmap = MOI.copy_to(model.optimizer, model.model_cache, copy_names=false)
131       model.state = ATTACHED_OPTIMIZER
132       # MOI does not define the type of index_map, so we have to copy it into a

About to run: (tuple)(false)
1|debug >
In attach_optimizer(model) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Utilities/cachingoptimizer.jl:128
128       @assert model.state == EMPTY_OPTIMIZER
129       # We do not need to copy names because name-related operations are handled by `m.model_cache`
130       indexmap = MOI.copy_to(model.optimizer, model.model_cache, copy_names=false)
131       model.state = ATTACHED_OPTIMIZER
132       # MOI does not define the type of index_map, so we have to copy it into a

About to run: (NamedTuple{(:copy_names,),T} where T<:Tuple)((false,))
1|debug >
In attach_optimizer(model) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Utilities/cachingoptimizer.jl:128
128       @assert model.state == EMPTY_OPTIMIZER
129       # We do not need to copy names because name-related operations are handled by `m.model_cache`
130       indexmap = MOI.copy_to(model.optimizer, model.model_cache, copy_names=false)
131       model.state = ATTACHED_OPTIMIZER
132       # MOI does not define the type of index_map, so we have to copy it into a

About to run: (Core.kwfunc)(MathOptInterface.copy_to)
1|debug >
In attach_optimizer(model) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Utilities/cachingoptimizer.jl:128
128       @assert model.state == EMPTY_OPTIMIZER
129       # We do not need to copy names because name-related operations are handled by `m.model_cache`
130       indexmap = MOI.copy_to(model.optimizer, model.model_cache, copy_names=false)
131       model.state = ATTACHED_OPTIMIZER
132       # MOI does not define the type of index_map, so we have to copy it into a

About to run: (getproperty)(<suppressed 28148 bytes of output>, :optimizer)
1|debug >
In attach_optimizer(model) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Utilities/cachingoptimizer.jl:128
128       @assert model.state == EMPTY_OPTIMIZER
129       # We do not need to copy names because name-related operations are handled by `m.model_cache`
130       indexmap = MOI.copy_to(model.optimizer, model.model_cache, copy_names=false)
131       model.state = ATTACHED_OPTIMIZER
132       # MOI does not define the type of index_map, so we have to copy it into a

About to run: (getproperty)(<suppressed 28148 bytes of output>, :model_cache)
1|debug >
In attach_optimizer(model) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Utilities/cachingoptimizer.jl:128
128       @assert model.state == EMPTY_OPTIMIZER
129       # We do not need to copy names because name-related operations are handled by `m.model_cache`
130       indexmap = MOI.copy_to(model.optimizer, model.model_cache, copy_names=false)
131       model.state = ATTACHED_OPTIMIZER
132       # MOI does not define the type of index_map, so we have to copy it into a

About to run: (<suppressed 52 bytes of output>)((copy_names = false,), MathOptInterface.copy_to, <suppressed 16897 bytes of output>, <suppressed 10650 bytes of output>)
1|debug > s
In #copy_to(, , mock, src) at none:0
?
About to run: (pairs)((copy_names = false,))
1|debug > nc
In #copy_to(, , mock, src) at none:0
?
About to run: (MathOptInterface.Bridges.#copy_to#1)(Base.Iterators.Pairs(:copy_names => 0), MathOptInterface.copy_to, <suppressed 16897 bytes of output>, <suppressed 10650 bytes of output>)
1|debug > s
In #copy_to#1(kws, , mock, src) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Bridges/bridgeoptimizer.jl:91
91      MOIU.automatic_copy_to(mock, src; kws...)
92  end
93  function MOIU.supports_default_copy_to(b::AbstractBridgeOptimizer,

About to run: (NamedTuple)()
1|debug > nc
In #copy_to#1(kws, , mock, src) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Bridges/bridgeoptimizer.jl:91
91      MOIU.automatic_copy_to(mock, src; kws...)
92  end
93  function MOIU.supports_default_copy_to(b::AbstractBridgeOptimizer,

About to run: (merge)(NamedTuple(), Base.Iterators.Pairs(:copy_names => 0))
1|debug >
In #copy_to#1(kws, , mock, src) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Bridges/bridgeoptimizer.jl:91
91      MOIU.automatic_copy_to(mock, src; kws...)
92  end
93  function MOIU.supports_default_copy_to(b::AbstractBridgeOptimizer,

About to run: (isempty)((copy_names = false,))
1|debug >
In #copy_to#1(kws, , mock, src) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Bridges/bridgeoptimizer.jl:91
91      MOIU.automatic_copy_to(mock, src; kws...)
92  end
93  function MOIU.supports_default_copy_to(b::AbstractBridgeOptimizer,

About to run: (Core.kwfunc)(MathOptInterface.Utilities.automatic_copy_to)
1|debug >
In #copy_to#1(kws, , mock, src) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Bridges/bridgeoptimizer.jl:91
91      MOIU.automatic_copy_to(mock, src; kws...)
92  end
93  function MOIU.supports_default_copy_to(b::AbstractBridgeOptimizer,

About to run: (<suppressed 72 bytes of output>)((copy_names = false,), <suppressed 44 bytes of output>, <suppressed 16897 bytes of output>, <suppressed 10650 bytes of output>)
1|debug > s
In #automatic_copy_to(, , dest, src) at none:0
?
About to run: (haskey)((copy_names = false,), :copy_names)
1|debug > nc
In #automatic_copy_to(, , dest, src) at none:0
?
About to run: (getindex)((copy_names = false,), :copy_names)
1|debug >
In #automatic_copy_to(, , dest, src) at none:0
?
About to run: (isa)(false, Bool)
1|debug >
In #automatic_copy_to(, , dest, src) at none:0
?
About to run: (Core.apply_type)(NamedTuple, (:copy_names,))
1|debug >
In #automatic_copy_to(, , dest, src) at none:0
?
About to run: (Base.structdiff)((copy_names = false,), NamedTuple{(:copy_names,),T} where T<:Tuple)
1|debug >
In #automatic_copy_to(, , dest, src) at none:0
?
About to run: (pairs)(NamedTuple())
1|debug >
In #automatic_copy_to(, , dest, src) at none:0
?
About to run: (isempty)(Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}())
1|debug >
In #automatic_copy_to(, , dest, src) at none:0
?
About to run: (<suppressed 48 bytes of output>)(false, <suppressed 44 bytes of output>, <suppressed 16897 bytes of output>, <suppressed 10650 bytes of output>)
1|debug > s

I'm not really sure what the best thing to do here is. Perhaps we should by default automatically skip any builtins (and maybe other calls inserted by the kwarg lowering). At the very least, this transcript shows that this causes some problems.

[KristofferC]

We already try to skip the "kwprep" (the setup before a kwarg call) but perhaps we miss some cases.

This would be easier to start working on if the function (ideally a MWE) was available to step through.

[timholy]

https://github.com/JuliaDebug/JuliaInterpreter.jl/blob/c3854f57fbedf65b16b788ab12a7f41c9d0f1230/src/commands.jl#L259-L322

[pfitzseb]

You sure you're on the latest versions of Debugger and JuliaInterpreter, @Keno? The REPL snippet looks like it's from a pretty old version.

[KristofferC]

Yeah, it doesn't show the line number indicator which was added quite a while back.

[pfitzseb]

Also prints the ? instead of the CodeInfo, which shouldn't happen after #22.

[Keno]

Indeed, looks like my check out was old in this project. Will try again with master.

[Keno]

Same stepping behavior on master. FWIW, the thing I'm stepping through is JuliaLang/julia#31439 (step into optimize!, optimize!, attach_optimizer, and that's where the call to copy_to is that starts above transcript).

[timholy]

Could you get to the beginning of the trouble and then hit L to display the lowered code, then paste here?

[Keno]

In attach_optimizer(model) at /home/keno/.julia/packages/MathOptInterface/C3lip/src/Utilities/cachingoptimizer.jl:128
  6  └──       goto #3 if not %4
  7  2 ─       goto #4
  8  3 ─ %7  = (AssertionError)("model.state == EMPTY_OPTIMIZER")
  9  └──       (throw)(%7)
>10      @ /home/keno/.julia/packages/MathOptInterface/C3lip/src/Utilities/cachingoptimizer.jl:130 within `attach_optimizer'
 11  4 ┄ %9  = (:copy_names,)
 12  │   %10 = (Core.apply_type)(NamedTuple, %9)
 13  │   %11 = (tuple)(false)
 14  │   %12 = (%10)(%11)

[timholy]

A little farther forward please. You're starting to match this pattern but we need to see 15 and 16.

[Keno]

 15  │   %13 = MathOptInterface.copy_to
>16  │   %14 = (Core.kwfunc)(%13)
 17  │   %15 = MathOptInterface.copy_to
 18  │   %16 = (getproperty)(model, :optimizer)
 19  │   %17 = (getproperty)(model, :model_cache)
 20  │         indexmap = (%14)(%12, %15, %16, %17)

[timholy]

OK, that's the problem; the kwfunc call is offset by 5 rather than 4 from the tuple construction. I wish I knew what the rules were and didn't just have to figure this stuff out from failures.

I will fix. Can I ask if fixing JuliaLang/julia#31073 is on your radar? (It was caused by your PR JuliaLang/julia#28707.) We've come up with a workaround for Base and the stdlibs, but any package that explicitly ccalls memcpy will crash, and there's nothing we can do about that.