VSCode debugger stops in a macro without breakpoint

Hello,

I'm using @einsum from the Tullio package, https://github.com/mcabbott/Tullio.jl, and I get an unexpected behavior when debugging a Julia program in VSCode. The debugger always stops at the same line in macro.jl, a file from Tullio, without having set a break point there.

Say, I have a program like

using Tullio: @einsum

v = [1; 2; 3];
w = [4; 5; 6];       # <== 1st breakpoint
@einsum a[i,j] := v[i] * w[j]; 
println("a=", a); 
b  = 4 * a;          # <== 2nd breakpoint 
println("b=", b);

After I start the debugger with F5, VSCode shows correct behavior at the 1st breakpoint, see first snapshot.
When I continue with F5 the debugger stops in line 826 of macro.jl, a file of Tullio, see 2nd snapshot.
Another continue with F5 leads me to the 2nd BP, see 3rd snapshot.
In case I have several @einsum statements the debugger stops several times always in macro.jl at line 826. That can be disturbing if I want to debug code after many @einsum statements.

I discussed this behavior with the author of Tullio, @mcabbott, and we think that it is a problem of VSCode.
With Base.remove_linenums!(@macroexpand1 @einsum a[i,j] := v[i] * w[j]) one can check the code generated by @einsum.
See mcabbott/Tullio.jl#33.

Any hints?

Ralf

After starting the debugger with F5 at the 1st breakpoint (as expected)

Continue F5 and the debugger stops in macro.jl (unexpected behavior)

Another Continue F5 and the debugger stops at the 2nd break point (as expected)

[pfitzseb]

Can you try putting the code in a function and calling that?

Also putting the few lines in a function leads to the unexpected behavior of stopping in macro.jl. Using @einsum in functions is my usual use case.

using Tullio: @einsum

function test()
    v = [1; 2; 3];
    w = [4; 5; 6];       # <== 1st breakpoint
    @einsum a[i,j] := v[i] * w[j]; 
    println("a=", a); 
    b  = 4 * a;          # <== 2nd breakpoint 
    println("b=", b); 
end
test()

[hs-ye]

I can confirm this happens when using the built in logging macros, @info and @warn etc.
Should be very easy to reproduce, just try running any of the below functions in the debugger

function test() # breaks on @info
    @info "test info"
    println("test print")
    @warn "test warn"
end

function test2() # breakpoint works fine
    println("test print")
end

test() will throw exception in the debugger, but test2() breaks just fine
Screenshot below: Pausing on the @info inside test, stepping to the next will produce an error

Error thrown by the @info macro of logging.jl

/Applications/Julia-1.5.app/Contents/Resources/julia/share/julia/base/logging.jl

[aviatesk]

my hunch is these problems are upstream ones, especially this one

[whubbla]

Hello,

I am experiencing more-or-less the same thing (I'm seeing now it relates to macros), though it happens to me when I'm using the package ODBC.jl, which uses macros; I'm not trying to use any of my own macros. My workaround for now is to debug in "Compiled Mode", which fixes the issue of breaking in macros included in packages where I don't want to break inside of the package, but this workaround is bothersome when I'm using multiple modules of my own and then can't set breakpoints in anything other than the main module.

What happens is that I get stuck in an endless loop of trying to log something (so, no amount of stepping out or over or continuing gets me back to my code) arising from some code in ODBC.jl.

Specifically, I run this code:
conn = ODBC.Connection(dsnname)

where dsnname is the name of a dsn I set up in Windows. I'm using Windows login, so no user name or password.

When it enters a function in ODBC called driverconnect(.), it goes to logging, i.e., the next thing up in the call stack is in base\logging.jl, and that's where my stepping line gets stuck when I'm trying to debug.

function driverconnect(connstr)
    dbc = Handle(SQL_HANDLE_DBC, ODBC_ENV[])
    @checksuccess dbc SQLDriverConnect(dbc, connstr)
    return dbc
end

Somehow I get stuck in this:

# Generate code for logging macros
function logmsg_code(_module, file, line, level, message, exs...)

The issue appears to involve the macro @checksucces

Also, I should note that if I just run the script in the REPL or execute the whole file, it works fine, in the sense that I can create an ODBC connection to MS SQL Server and get data back.

[chilledgeek]

Any updates on this issue? I agree with @spcogg that I am seeing the same issue with the built in logging macros on my machine (Ubuntu 20.10 running on VirtualBox).

• Here's a simple code alongside a test script
  
• The test script runs fine in normal non-debug mode (even the Compiled Mode works).
• However, when debugging, it does not get to the breakpoint (line 7 on the left), as it seems to hang while trying to run line 6's @info ... and does not further proceed
  

[pfitzseb]

This bug will be fixed with the next release.

[pfitzseb]

The original bug is still present while anything related to the logging macros has been fixed.

[matthijsvk]

any updates on this?
The debugger still goes to logging.jl line 317, then to _invoded_shouldlog() when running this code with breakpoint on println:

function test() # breaks on @info
    @info "test info"
    println("test print")
    @warn "test warn"
end

[pfitzseb]

We haven't published a new release yet I think. The insiders build should definitely have the issue fixed though.