[Lldb-commits] [PATCH] D119963: [LLDB] Dump valid ranges of variables

[Jim Ingham via lldb-commits]

There was a question on the dev list a while ago about to print out all the valid ranges of a variable.  That's a useful bit of info if you're trying to figure out where you could break to actually see or change a variable's value in an optimized function.  I think that's the motivation for the change, so just showing the range that encompasses the address wouldn't satisfy that initial impetus for the change.

It is a little awkward to put it in "image lookup -va" however, and odder to have it be part of the printing of an Address.  It seems more like something you should ask a function?

Jim


> On Feb 17, 2022, at 12:57 AM, Pavel Labath via Phabricator <reviews at reviews.llvm.org> wrote:
> 
> labath added a comment.
> 
> I suppose one could ask the question whether this should be printing the entire set range of ranges, or just the one
> 
> 
> 
> ================
> Comment at: lldb/source/Commands/Options.td:960
>     "target modules.">;
> +  def target_modules_lookup_variables_ranges : Option<"show-variables-ranges", 
> +    "R">, GroupRange<1, 6>, Desc<"Dump valid ranges of variables (must be used "
> ----------------
> 
> 
> 
> ================
> Comment at: lldb/source/Commands/Options.td:961
> +  def target_modules_lookup_variables_ranges : Option<"show-variables-ranges", 
> +    "R">, GroupRange<1, 6>, Desc<"Dump valid ranges of variables (must be used "
> +    "in conjunction with --verbose">;
> ----------------
> A not widely known fact: It is possible to produce an long option without a short version by using an non-printable character in place of the short option. You can see and example of that in `disassemble --force`, defined in this file.
> 
> If we are considering this a niche option, then maybe we don't need to use a random character just to have a short version.
> 
> 
> ================
> Comment at: lldb/test/Shell/SymbolFile/DWARF/x86/debug_loc.s:16
> +# CHECK-LABEL: image lookup -v -a 0 -R
> +# CHECK: Variable: {{.*}}, name = "x0", type = "int", location = DW_OP_reg5 RDI, decl = , valid ranges =
> +# CHECK: Variable: {{.*}}, name = "x1", type = "int", location = <empty>, decl = , valid ranges =
> ----------------
> I don't think that the information itself is out of place here, but I think it can get confusing when one sees it like this, next to the `location` field. So, if we had an output like `location = RDI, valid ranges = [0, 20)`, I think one would/could read it as "the variable is in RDI from PC=0 to PC=20", but what it _really_ is telling you is "the variable (or its value -- I'm not sure) exists between PC=0 and PC=20, *and* for the exact PC that you are querying (0 in this case), it's value happens to be in RDI".
> 
> We could try to come up with a way to make this less confusing, but maybe we could just sidestep this problem and just print this information in the `image dump symfile` output.
> 
> 
> ================
> Comment at: lldb/test/Shell/SymbolFile/DWARF/x86/debug_loc.s:27
> # CHECK:   Function{
> # CHECK:     Variable{{.*}}, name = "x0", {{.*}}, scope = parameter, location =
> # CHECK-NEXT:  [0x0000000000000000, 0x0000000000000001): DW_OP_reg5 RDI
> ----------------
> If you printed the information here, then there would be no (or less) confusion as the location field contains explicit range information. Additionally, the `image dump symfile` output is so verbose, than an extra field does not make a difference.
> 
> (I'm sorry, I know it was I who suggested using `image lookup`, but I did not realize you would be adding an extra argument because of that.)
> 
> 
> Repository:
>  rG LLVM Github Monorepo
> 
> CHANGES SINCE LAST ACTION
>  https://reviews.llvm.org/D119963/new/
> 
> https://reviews.llvm.org/D119963
>