This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: egcs build report



  In message <Pine.LNX.3.96.980619105856.5081A-100000@bhalpha1.nibh.go.jp>you write:
  > >I don't want to suddenly get lots of mail from folks who can't debug
  > >their code (because someone installed a libstdc++ with no debug symbols).
  > 
  > Actually, this is a good point.  Although I don't know why c++ debugging
  > requires library symbols (I can debug my C code even though I can't step
  > into strcpy (why would I want to?)).
For example, you can't look at the internals of any classes defined in 
the library.

As someone that used to put together binary releases, I got my hand
slapped more than once for building the C++ runtime libraries without
debug symbols :-)

  > But the compiler itself surely
  > doesn't need -g to run.  Perhaps some selectivity is called for.
Users can change BOOT_CFLAGS to modify which are flags are used during 
the build.  They can also install stripped (or just debug stripped)
binaries if they desire (it's more complicated, but possible).

  > As long as we're on this, could somebody please satisfy my curiosity about
  > this: does -g impact the performance of -O?  In some compilers, -g turns
  > -O off, and that's mostly why I don't use it (unless I specifically want
  > to run gdb).  But what's the story with gcc/egcs?
-g has no impact on the generated code.  This has been a policy set in
stone for many years.  If you ever encounter a case where -g changes the
generated code, it would be considered a bug.


jeff


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]