@@ -48,18 +48,17 @@ underspecified for some variations of symbol file deployment. The protocol
48
48
itself is quite simple: query an HTTP server with the path
49
49
` buildid/{.note.gnu.build-id hash}/debuginfo ` or
50
50
` buildid/{.note.gnu.build-id hash}/executable ` to acquire "symbol data" or "the
51
- executable". Where there is lack of clarity, I prefer requesting ` debuginfo `
52
- first, then falling back to ` executable ` (Scenarios #1 & #2 ). For Scenario #5 ,
53
- I've chosen to expect the stripped (i.e. not full) executable, which contains a
54
- number of sections necessary to correctly symbolicate will be hosted from the
55
- ` executable ` API. Depending upon how Debuginfod hosting services choose to
56
- support ` .dwp ` paired with stripped files, these assumptions may need to be
57
- revisited.
51
+ executable". Where there is lack of clarity, ` debuginfo ` is requested first,
52
+ falling back to ` executable ` (Scenarios #1 & #2 ). For Scenario #5 , the stripped
53
+ (i.e. not full) executable, which contains a number of sections necessary to
54
+ correctly symbolicate, is hosted from the ` executable ` API. Depending upon how
55
+ Debuginfod hosting services choose to support ` .dwp ` paired with stripped files,
56
+ these assumptions may need to be revisited.
58
57
59
- I've also chosen to simply treat the ` .dwp ` file as ` debuginfo ` and the
60
- "only-keep-debug" stripped binary as ` executable ` . This scenario doesn't appear
61
- to work at all in GDB. Supporting it how I did seems more straight forward than
62
- trying to extend the protocol. The protocol _ does _ support querying for section
63
- contents by name for a given build ID, but adding support for that in LLDB
64
- looks...well beyond my current capability (and LLVM's Debuginfod library doesn't
65
- support it at this writing, anyway ).
58
+ The ` .dwp ` file is treated as ` debuginfo ` , and the "only-keep-debug" stripped
59
+ binary is treated as ` executable ` . This scenario does not appear to work at all
60
+ in GDB. It seems more straightforward to support it in this manner, rather than
61
+ attempting to extend the protocol. While the protocol does support querying for
62
+ section contents by name for a given build ID, support for that capability in
63
+ LLDB is not implemented (and LLVM's Debuginfod library does not support it at
64
+ this time ).
0 commit comments