* Roland McGrath <rol...@redhat.com> wrote:
> > ( Note, Linus might still insist on a total replacement, if he
> > finds the #ifdef approach too ugly. I dont talk for him and he
> > is usually much pickier than me. )
>
> In a previous round of review, hch objected to
> CONFIG_UTRACE_PTRACE. I think we are all in agreement that the
> eventual right place will be only one ptrace implementation, and
> that being the one based on a clean framework. It's not very
> clear to me which different incremental paths to get there
> different people have in mind or why.
>
> Everyone agrees #ifdef for two implementations is ugly. It's a
> transitional stage, so to me it seems quite tolerable knowing that
> it will be cleaned up eventually. It buys two things: 1. getting
> utrace in sooner, worked on faster, and made better soon; 2. given
> that, risk mitigation for everyone not interested in working with
> utrace.
The problem for upstream is, if it goes in ugly and everyone gets
what they wanted they often go and chase other targets. Especially
if it's such an external-looking and external-thinking project as
SystemTap. Such incidents happened frequently enough to upstream to
become a primary worry.
[ For example: you promised proper x86 CFI annotations macro design
one year ago to Linus and me, in exchange for me not removing the
ugly ones. I already had the removal patches done and committed at
that stage and reverted them after that. The ugly CFI stuff is
still there today and it's all bitrotting nicely ;-) ]
And there's a slam-dunk counter argument: "ptrace is ugly enough
already, we dont need another 'temporary' layer".
So 'temporary ugliness' is being frowned upon. Ugliness might be
taken from trusted parties in well-argued cases but it is still
exceedingly rare.
Ingo
