Re: clang-8.0 very strict -- can we think about changing the clang compiler list order?

2019-09-28 Thread Ken Cunningham
Apple does a lot in it's clang fork to relax the restrictiveness. 

Seriously, set your default compilers to what 10.6.8 has to live with; at least 
share in it with us!

But I can see I'm outvoted here. My great dream of no older systems tickets 
dies.

:)

K 

> On Sep 28, 2019, at 01:14, Chris Jones  wrote:
> 
> 
> 
>> On 27 Sep 2019, at 3:06 pm, Ken Cunningham  
>> wrote:
>> 
>> I think it would be fair to invite everyone to try setting the default 
>> compiler list for a week on your own systems to our current macports clang 
>> progression!
>> 
>> Compilers are always ahead of the codebase by a couple of years.
>> 
>> What are the clang-using Linux distros defaulting, I wonder? That might tell 
>> us where they feel the codebase is presently...
> 
> Our point of reference should be the Xcode clang versions. Most ports build 
> fine on 10.14 using the system clang there, so what upstream version does 
> this equate to , roughly ? My guess based on release times would be somewhere 
> are clang 7. 
> 
> I am with Jeremy here, we need to start pushing away from clang 5.
> 
> Chris
> 
>> 
>> Ken
>> 
> 


Re: clang-8.0 very strict -- can we think about changing the clang compiler list order?

2019-09-28 Thread Chris Jones



> On 27 Sep 2019, at 3:06 pm, Ken Cunningham  
> wrote:
> 
> I think it would be fair to invite everyone to try setting the default 
> compiler list for a week on your own systems to our current macports clang 
> progression!
> 
> Compilers are always ahead of the codebase by a couple of years.
> 
> What are the clang-using Linux distros defaulting, I wonder? That might tell 
> us where they feel the codebase is presently...

Our point of reference should be the Xcode clang versions. Most ports build 
fine on 10.14 using the system clang there, so what upstream version does this 
equate to , roughly ? My guess based on release times would be somewhere are 
clang 7. 

I am with Jeremy here, we need to start pushing away from clang 5.

Chris

> 
> Ken
> 



Re: clang-8.0 very strict -- can we think about changing the clang compiler list order?

2019-09-27 Thread Jeremy Huddleston Sequoia via macports-dev



> On Sep 26, 2019, at 21:46, Ken Cunningham  
> wrote:
> 
>> FYI clang-8.0, new default on 10.6.8, is very strict
>> 
>> appears will have to add to blacklist on many ports
> 
> 
> So as much as I would have put clang-5.0 first as the default compiler if the 
> system clang won't work, I understand that as some payback for all this older 
> systems support effort, it's nice to get a heads-up regarding which ports are 
> going to break with future Xcode updates, and defaulting to clang-8.0 does 
> provide that.
> 
> But the way it is now is not ideal.
> 
> When a clang-8.0 build fails, we fall back to clang-7.0, test that, then 
> clang-6.0, test that, then clang-5.0 (which builds everything).
> 
> What this means for a dev is a lot of rebuilding. Like qt4-mac. 
> 
> I know clang-8.0 fails, and clang-5.0 works. It doesn't make a lot of sense 
> to me to spend hours and hours figuring out where the real cutoff is. I can 
> just blacklist {clang-[6-9].0} , and move on, but I don't know if clang 6 can 
> build it, so that's not idea.

No, we should not blacklist compilers just because we don't want to test them.
> 
> What it means for a user is eventually four different clangs installed (5, 6, 
> 7, 8), all of them defaulted for different ports.
> 
> How about we set the list like this:
> 
> clang-8.0 -> clang-5.0 -> clang7.0 -> clang-6.0.

clang-5.0 is likely on the chopping block soon as it is also a couple years 
old.  We should really fix anything that fails to build with clang-8.

> That way, we will only have to blacklist clang-8.0 when it fails, and the 
> port will 99% of the time build with clang-5.0, and you're done.
> 
> And the end user will perhaps only have two clangs, 8.0 and 5.0, and we can 
> soon get rid of the others (6,7) which look to be obsolete soon enough.

Why would we want to keep 5.0 around and obsolete 6 and 7?  5.0 provides no 
bootstrapping benefit.

> Best,
> 
> Ken



Re: clang-8.0 very strict -- can we think about changing the clang compiler list order?

2019-09-27 Thread Ken Cunningham
I think it would be fair to invite everyone to try setting the default compiler 
list for a week on your own systems to our current macports clang progression!

Compilers are always ahead of the codebase by a couple of years.

What are the clang-using Linux distros defaulting, I wonder? That might tell us 
where they feel the codebase is presently...

Ken



Re: clang-8.0 very strict -- can we think about changing the clang compiler list order?

2019-09-27 Thread Chris Jones

Hi,

I would not be particularly in favour of changing things as you suggest. 
For me the defaults as they are are correct. They should start with the 
newest clang available and work back, in order.


For specific ports that have issues, either fix the port to work with a 
recent clang, or if not practical you can always explicitly set the 
fallback or whitelist to be whatever ports you want, in whatever order. 
I would prefer to hacks like this specific to each port that needs them, 
not hidden away in base, just to make it clear when looking at the port 
file itself what is going on.


Also personally I have no issue with a mixture of macports clangs being 
used by ports, so don't see the issue with a user having to have a 
number installed. Most users get prebuilt binaries anyway.


Chris

On 27/09/2019 5:46 am, Ken Cunningham wrote:

FYI clang-8.0, new default on 10.6.8, is very strict

appears will have to add to blacklist on many ports



So as much as I would have put clang-5.0 first as the default compiler if the 
system clang won't work, I understand that as some payback for all this older 
systems support effort, it's nice to get a heads-up regarding which ports are 
going to break with future Xcode updates, and defaulting to clang-8.0 does 
provide that.

But the way it is now is not ideal.

When a clang-8.0 build fails, we fall back to clang-7.0, test that, then 
clang-6.0, test that, then clang-5.0 (which builds everything).

What this means for a dev is a lot of rebuilding. Like qt4-mac.

I know clang-8.0 fails, and clang-5.0 works. It doesn't make a lot of sense to 
me to spend hours and hours figuring out where the real cutoff is. I can just 
blacklist {clang-[6-9].0} , and move on, but I don't know if clang 6 can build 
it, so that's not idea.

What it means for a user is eventually four different clangs installed (5, 6, 
7, 8), all of them defaulted for different ports.

How about we set the list like this:

clang-8.0 -> clang-5.0 -> clang7.0 -> clang-6.0.

That way, we will only have to blacklist clang-8.0 when it fails, and the port 
will 99% of the time build with clang-5.0, and you're done.

And the end user will perhaps only have two clangs, 8.0 and 5.0, and we can 
soon get rid of the others (6,7) which look to be obsolete soon enough.

Best,

Ken



Re: clang-8.0 very strict -- can we think about changing the clang compiler list order?

2019-09-26 Thread Ken Cunningham
> FYI clang-8.0, new default on 10.6.8, is very strict
> 
> appears will have to add to blacklist on many ports


So as much as I would have put clang-5.0 first as the default compiler if the 
system clang won't work, I understand that as some payback for all this older 
systems support effort, it's nice to get a heads-up regarding which ports are 
going to break with future Xcode updates, and defaulting to clang-8.0 does 
provide that.

But the way it is now is not ideal.

When a clang-8.0 build fails, we fall back to clang-7.0, test that, then 
clang-6.0, test that, then clang-5.0 (which builds everything).

What this means for a dev is a lot of rebuilding. Like qt4-mac. 

I know clang-8.0 fails, and clang-5.0 works. It doesn't make a lot of sense to 
me to spend hours and hours figuring out where the real cutoff is. I can just 
blacklist {clang-[6-9].0} , and move on, but I don't know if clang 6 can build 
it, so that's not idea.

What it means for a user is eventually four different clangs installed (5, 6, 
7, 8), all of them defaulted for different ports.

How about we set the list like this:

clang-8.0 -> clang-5.0 -> clang7.0 -> clang-6.0.

That way, we will only have to blacklist clang-8.0 when it fails, and the port 
will 99% of the time build with clang-5.0, and you're done.

And the end user will perhaps only have two clangs, 8.0 and 5.0, and we can 
soon get rid of the others (6,7) which look to be obsolete soon enough.

Best,

Ken

clang-8.0 very strict

2019-09-24 Thread Ken Cunningham
FYI clang-8.0, new default on 10.6.8, is very strict

appears will have to add to blacklist on many ports
or otherwise surgerize code/standard setting — eventually will have to happen 
but all at once is daunting

downgrading default compiler on 10.6.8 to clang-5.0 gets the rev-upgrade done

once system is restored and working again will go back and try to start in on 
the clang-8.0 blacklisting

K