Handle DW_TAG_partial_unit

[calixteman]

I'm trying to use dump_syms which depends on symbolic which itself depends on gimli to dump breakpad symbols for iris_dri.so-20.1.1-1.fc32.x86_64.debug.
I get neither DW_TAG_subprogram nor DW_TAG_inlined_subroutine, all the units are either empty or erroneous.
I llvm-dwarf dump the file:
The first 54 CU looks like:

0x00000000: Compile Unit: length = 0x00000013 version = 0x0004 abbr_offset = 0x2842 addr_size = 0x08 (next unit at 0x00000017)

0x0000000b: DW_TAG_partial_unit

0x0000000c:   DW_TAG_imported_unit
                DW_AT_import	(<alt 0x2a6a57>)

0x00000011:   DW_TAG_imported_unit
                DW_AT_import	(<alt 0x2a6a73>)

0x00000016:   NULL

These ones look empty.
And all the next CU start with something like:

0x000000cf: Compile Unit: length = 0x0000001b version = 0x0004 abbr_offset = 0x15987 addr_size = 0x08 (next unit at 0x000000ee)

0x000000da: DW_TAG_partial_unit
              DW_AT_stmt_list	(0x00000000)
              DW_AT_comp_dir	()

0x000000e3:   DW_TAG_imported_unit
                DW_AT_import	(<alt 0xb>)

0x000000e8:   DW_TAG_imported_unit
                DW_AT_import	(<alt 0x2a6b29>)

0x000000ed:   NULL

they look like erroneous.
I greped on TAG_partial_unit on the gimli source and I got almost nothing so maybe it's because gimli doesn't handle partial unit.

[bjorn3]

What is the llvm-dwarfdump output?

[calixteman]

The dump for .debug_info is ~700Mb so it's why I sumed it up in my comment 0.

[jrmuizel]

It looks this is a regression.
With 3015ebe

$ cargo run --release --example dwarfdump /tmp/iris_dri.so-20.1.1-1.fc32.x86_64.debug | grep subprogram | wc
    Finished release [optimized] target(s) in 0.03s
     Running `target/release/examples/dwarfdump /tmp/iris_dri.so-20.1.1-1.fc32.x86_64.debug`
   33444  100332 1278020

With master

$ cargo run --release --example dwarfdump /tmp/iris_dri.so-20.1.1-1.fc32.x86_64.debug | grep subprogram | wc
    Finished release [optimized] target(s) in 0.03s
     Running `target/release/examples/dwarfdump /tmp/iris_dri.so-20.1.1-1.fc32.x86_64.debug`
       0       0       0

[jrmuizel]

I'll run a bisection.

[jrmuizel]

It looks like this was caused by 9dabba0

[mstange]

I've uploaded the file (gzipped) here: iris_dri.so-20.1.1-1.fc32.x86_64.debug.gz

[jrmuizel]

More specifically with that change DwarfUnit returns Error because dwarf.attr_string(&unit, val) on comp_dir does.

[jrmuizel]

diff --git a/src/read/dwarf.rs b/src/read/dwarf.rs
index 04fd013..eeb1a74 100644
--- a/src/read/dwarf.rs
+++ b/src/read/dwarf.rs
@@ -601,25 +601,25 @@ impl<R: Reader> Unit<R> {
                         if let AttributeValue::DebugRngListsBase(base) = attr.value() {
                             unit.rnglists_base = base;
                         }
                     }
                     _ => {}
                 }
             }
         }
 
         unit.name = match name {
-            Some(val) => Some(dwarf.attr_string(&unit, val)?),
+            Some(val) => dwarf.attr_string(&unit, val).ok(),
             None => None,
         };
         unit.comp_dir = match comp_dir {
-            Some(val) => Some(dwarf.attr_string(&unit, val)?),
+            Some(val) => dwarf.attr_string(&unit, val).ok(),
             None => None,
         };
         unit.line_program = match line_program_offset {
             Some(offset) => Some(dwarf.debug_line.program(
                 offset,
                 unit.header.address_size(),
                 unit.comp_dir.clone(),
                 unit.name.clone(),
             )?),
             None => None,

fixes it.