Re: 32-bit APIs with 64-bit "data"

2010-10-29 Thread Leif Hedstrom
On 10/29/2010 08:12 AM, Patrick McManus wrote: On Thu, 2010-10-28 at 08:38 -0600, Leif Hedstrom wrote: while looking at some issues we're having with InkAPI plugins, the issue I saw the message yesterday on the changing of the plugin A{BP}I in a non-compatible way in order to support 64 bit l

Re: 32-bit APIs with 64-bit "data"

2010-10-29 Thread Patrick McManus
On Thu, 2010-10-28 at 08:38 -0600, Leif Hedstrom wrote: > while looking at some issues we're having with InkAPI plugins, the issue I saw the message yesterday on the changing of the plugin A{BP}I in a non-compatible way in order to support 64 bit lengths. (i.e. option 2). I hope if you do that

Re: 32-bit APIs with 64-bit "data"

2010-10-28 Thread Mark Scifres
Me too +1 - Original Message - From: Theo Schlossnagle To: dev@trafficserver.apache.org Sent: Thu Oct 28 19:45:10 2010 Subject: Re: 32-bit APIs with 64-bit "data" I think the public adoption is really low, so the pain involved with breaking API compatibility is relativel

Re: 32-bit APIs with 64-bit "data"

2010-10-28 Thread Theo Schlossnagle
I think the public adoption is really low, so the pain involved with breaking API compatibility is relatively low. So, go for it. #2 +1 On Thu, Oct 28, 2010 at 4:14 PM, Alan M. Carroll < a...@network-geographics.com> wrote: > Now that John has bravely stepped up, I will agree with him on select

Re: 32-bit APIs with 64-bit "data"

2010-10-28 Thread Alan M. Carroll
Now that John has bravely stepped up, I will agree with him on selecting on option #2.

Re: 32-bit APIs with 64-bit "data"

2010-10-28 Thread John Plevyak
resend On 10/28/10 09:57, Leif Hedstrom wrote: My gut feeling is though, if we're willing to break API, we should just make these APIs 64-bit :). And even the size_t solution would break ABIs, i.e. plugins compiled with the old APIs will no longer work if they upgrade TS (on a 64-bit system).

Re: 32-bit APIs with 64-bit "data"

2010-10-28 Thread John Plevyak
On 10/28/10 09:57, Leif Hedstrom wrote: My gut feeling is though, if we're willing to break API, we should just make these APIs 64-bit :). And even the size_t solution would break ABIs, i.e. plugins compiled with the old APIs will no longer work if they upgrade TS (on a 64-bit system). On re

Re: 32-bit APIs with 64-bit "data"

2010-10-28 Thread Leif Hedstrom
On 10/28/2010 10:45 AM, John Plevyak wrote: This is based on experience with FreeBSD vs Linux where FreeBSD took the route that they would upgrade all values to be 64-bit for 64-bit platforms whereas Linux has taken the #3 route in places which has lead to lots of confusion and subtle problems

Re: 32-bit APIs with 64-bit "data"

2010-10-28 Thread John Plevyak
My feeling is that #2 is the way to go. This is based on experience with FreeBSD vs Linux where FreeBSD took the route that they would upgrade all values to be 64-bit for 64-bit platforms whereas Linux has taken the #3 route in places which has lead to lots of confusion and subtle problems. On