On 22 Mar 2012, at 00:00, Steven Bosscher wrote:

On Thu, Mar 22, 2012 at 12:53 AM, Iain Sandoe
<develo...@sandoe-acoustics.co.uk> wrote:

conceptually, the issue is that there are multiple sets of built-ins
(potentially, one set for each runtime, and the sets are of different
sizes). Thus, it's not just a case of turning these into regular built-ins, without some mechanism to cater for overloading or presence/absence of
particular ones.

I don't understand this. We're committed to one runtime per
compilation, right? If so, then we should create only one of the sets
at any time.

Yes, that's true [notwithstanding an erroneous invocation of LTO with mixed inputs, which we should detect by alternate means]

but ...
.. don't the indices for built-ins need to be constant?
(maybe that means that we'd just allocate as many as needed to cover all runtimes?)
.. or, otherwise, how does LTO know which set to invoke?

... sorry, I think my observation about pr24777 has pushed this thread off course - perhaps we'd be better putting this in the PR thread?


Reply via email to