[Bug mi/15806] Some fields in async MI events get escaped twice

["asmwarrior at gmail dot com" <sourceware-bugzilla at sourceware dot org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=15806

--- Comment #13 from asmwarrior <asmwarrior at gmail dot com> ---
(In reply to Simon Marchi from comment #12)
> Oops, yeah the fix was only for breakpoint-created, sorry about that.
Never mind.

> 
> Your comment made me realize that if a library path contains a quote ("),
> the result will be wrong. Example:
> 
> =library-loaded,id="/tmp/hello"you/libpopt.so.0",target-name="/tmp/hello"you/
> libpopt.so.0",host-name="/tmp/hello"you/libpopt.so.0",symbols-loaded="0",
> thread-group="i1"
> 
> To reproduce this, I created '/tmp/hello"you', copied some library in it and
> modified LD_LIBRARY_PATH so that my test program use this library instead of
> the system's version. You can see that the " between hello and you is not
> escaped, but should be.
Yes, look at the function body of mi_solib_loaded (struct so_list *solib). It
firstly use fprintf_unfiltered, which directly copy the contents to the stream
(such as solib->so_original_name, solib->so_original_name). So, it failed to
escape some special chars, such as the " in your file path.

> 
> I think that no escaping can be done reliably in gdb_flush. 
What does this sentence meansï You mean gdb_flush should only copy the contents
to the UI, it should not do escaping, right?


> There is no way
> to differentiate a quote that should not be escaped from one that should be
> escaped. Therefore, I think it should always be done at the moment where the
> content is created.

Correct, gdb_flush know nothing about whether the buffered stream need to be
escaped or not. (unless we write a parser to parse the whole stream, basically,
the stream has some format like xxxx="yyyyyy", the parser know whether it is
inside a string literal or not, but this way(direction) is surely wrong)

> 
> Also, using fprintf_unfiltered to output this is wrong, because the fields
> are never escaped. It would probably be better to use the ui_out_field_*
> functions everywhere. What do you think ?
Yes, I agree with you, the only way to do escaping correctly is call some
ui_out_field_* functions. But this need a lot of code change in GDB.

-- 
You are receiving this mail because:
You are on the CC list for the bug.