frame variable command, child objects, depth limits

Subprojects LLDB
#1

The frame variable command doesn’t seem to obey printing depth limits, instead having a ‘-D’ parameter that defaults to unlimited / infinity.

In the Dylan language implementation, we have objects that refer to themselves regularly (example is an empty list is a pair, head and tail, each of which points to the empty list). There are plenty of other situations where this happens.

I thought that perhaps I would just set a different default when loading my scripts, but there doesn’t seem to be a setting for the depth parameter to “frame variable”.

Some example output of what I’ve done can be seen: http://hastebin.com/nugasalavu.lua

Any advice on solving the issue with reference loops in child objects?

Thanks,

  • Bruce
#2

The depth limits only currently apply to pointers and references. Anything that is considered and instance will always have the entire contents printed.

Would this explain what you are seeing?

Greg

#3

The frame variable command doesn’t seem to obey printing depth limits, instead having a ‘-D’ parameter that defaults to unlimited / infinity.

Bruce,
I am not sure what “printing depth limits” you mean, if not the -D flag.
If that flag’s value is not honored, we should probably fix it. If you mean something else, can you elaborate a little bit more?

In the Dylan language implementation, we have objects that refer to themselves regularly (example is an empty list is a pair, head and tail, each of which points to the empty list). There are plenty of other situations where this happens.

Is this done via pointers? And do you expose them as such? If so, the —ptr-depth would be the flag to use

I thought that perhaps I would just set a different default when loading my scripts, but there doesn’t seem to be a setting for the depth parameter to “frame variable”.

No, but it would be interesting to have a way to set default arguments to commands - We don’t have that feature yet though

#4

The frame variable command doesn't seem to obey printing depth limits,
instead having a '-D' parameter that defaults to unlimited / infinity.

Bruce,
I am not sure what “printing depth limits” you mean, if not the -D flag.
If that flag’s value is not honored, we should probably fix it. If you
mean something else, can you elaborate a little bit more?

I meant the pointer depth limits aren't being obeyed (but I'm also not
using pointers). The explanation from Greg in his response to my email was
correct.

The -D flag does work.

In the Dylan language implementation, we have objects that refer to

themselves regularly (example is an empty list is a pair, head and tail,
each of which points to the empty list). There are plenty of other
situations where this happens.

Is this done via pointers? And do you expose them as such? If so, the
—ptr-depth would be the flag to use

Well, due to other issues (like the thread we have going on the commits
list), I'm unable to use pointers. I've modified our run-time to instead
represent Dylan values as uintptr_t so that I can have synthetic types for
them.

I thought that perhaps I would just set a different default when loading my

scripts, but there doesn't seem to be a setting for the depth parameter to
"frame variable".

No, but it would be interesting to have a way to set default arguments to
commands - We don’t have that feature yet though

It seems like just having a process or global setting here might be useful?

Right now, if we just run "frame variable" while debugging a Dylan app,
we're pretty much guaranteed to hang LLDB. We have to always pass a -D
flag.

- Bruce