Re: [patch] better dwarf2 debug info for inlined functions

[Mike Stump <mrs at apple dot com>]

On Jul 7, 2005, at 6:55 AM, Daniel Berlin wrote:
> You compile with debug info and optimization, and are going to say the
> debug info is too big?

I'll say it.  Our competition produces debug information around 16x  
smaller.  When one blows out the vm space of the linker, I mean  
address bits, not swap, things are kinda evil.

We see three technologies to fix this problem:

    1   Repository
    2   Compression
    3   Elimination of stuff the user doesn't need

I suspect we're going to gear up and start talking about these  
things.  To date, we have some patches that hook up debug information  
into -frepo, and probably plans to extend -frepo to -frepo=a.out and  
manage the repo files as a single unit for the final build product,  
this allow one to manage the answers given by the repo to be accurate  
the first time through compilation, instead of requiring recompiles  
to fix up earlier wrong answers.  The motivation for earlier right  
answers is that then one doesn't need to perform n link cycles, there  
n is the template depth (or debugging information depth).

Those opposed to 1 are invited to elaborate on their favorite  
solution.  A good solution would of course feature faster  
compilations, faster (5x) linking, smaller translation units (13x)...