On Wed, 12 Dec 2012, Iain Buclaw wrote:

> On 12 December 2012 17:29, Brad Anderson <eco@gnuk.net> wrote:
> > On Monday, 10 December 2012 at 00:34:33 UTC, Walter Bright wrote:
> >>
> >> It's time to do a release; to that end we should be working on tidying up the regressions.
> >>
> >> This will be the last official D1 release.
> >
> >
> > Just a heads up, GitHub has removed their Uploads feature[1].  Current uploads still work but this next release is going to need to either go back to the Digital Mars server like it was before or find a new home. I prefer the latter for speed reasons. Amazon S3 is fast and works well (it's what GitHub Uploads was backed by) and should be fairly affordable for the sizes we are talking about (I'd guess $3-4 per month).
> >
> > http://docs.amazonwebservices.com/AmazonS3/latest/gsg/GetStartedWithS3.html
> >
> >
> >
> > [1] https://github.com/blog/1302-goodbye-uploads
> 
> I can offer a server for that, hosted a datacentre in the UK.  If I recall correctly, the net link are behind a 100GB link, testing the net speed, I get 250MB/s upload speed to Europe, only 20MB/s upload to America though.
> 
> 
> Regards,
> -- 
> Iain Buclaw

Generous offer.  I've been meaning to build packaging into the auto-tester for both release builds and more frequent (nightly or maybe even every cycle) builds.  I was going to toss them into s3 with a cloudfront distribution in front of that.  Sounds like that needs to go higher on the todo list given github's recent changes.

The advantage of CloudFront is fast access all over the planet, though it has a cost associated with it.  I'll bear that until it starts to hurt.

On 12 December 2012 21:10, Brad Roberts <braddr@puremagic.com> wrote:
> On Wed, 12 Dec 2012, Iain Buclaw wrote:
>
>> On 12 December 2012 17:29, Brad Anderson <eco@gnuk.net> wrote:
>> > On Monday, 10 December 2012 at 00:34:33 UTC, Walter Bright wrote:
>> >>
>> >> It's time to do a release; to that end we should be working on tidying up the regressions.
>> >>
>> >> This will be the last official D1 release.
>> >
>> >
>> > Just a heads up, GitHub has removed their Uploads feature[1].  Current uploads still work but this next release is going to need to either go back to the Digital Mars server like it was before or find a new home. I prefer the latter for speed reasons. Amazon S3 is fast and works well (it's what GitHub Uploads was backed by) and should be fairly affordable for the sizes we are talking about (I'd guess $3-4 per month).
>> >
>> > http://docs.amazonwebservices.com/AmazonS3/latest/gsg/GetStartedWithS3.html
>> >
>> >
>> >
>> > [1] https://github.com/blog/1302-goodbye-uploads
>>
>> I can offer a server for that, hosted a datacentre in the UK.  If I recall correctly, the net link are behind a 100GB link, testing the net speed, I get 250MB/s upload speed to Europe, only 20MB/s upload to America though.
>>
>>
>> Regards,
>> --
>> Iain Buclaw
>
> Generous offer.  I've been meaning to build packaging into the auto-tester for both release builds and more frequent (nightly or maybe even every cycle) builds.  I was going to toss them into s3 with a cloudfront distribution in front of that.  Sounds like that needs to go higher on the todo list given github's recent changes.
>
> The advantage of CloudFront is fast access all over the planet, though it has a cost associated with it.  I'll bear that until it starts to hurt.


It's one of the perks of working in a UK-based cloud hosting company. I have been given the go ahead to build a server used as an auto-tester (initially intended specifically for GDC, though can throw in a DMD auto-tester too).  Though one of the downsides would be that if I were to leave, so would the site.


I also have another server host in the UK by Linode (this I pay for) - currently only used for gdcproject.org - this could be used as an EU based mirror for downloads.


Regards
-- 
Iain Buclaw

*(p < e ? p++ : p) = (c & 0x0f) + '0';

On Wednesday, 12 December 2012 at 07:01:14 UTC, Rob T wrote:
> On Wednesday, 12 December 2012 at 05:32:13 UTC, SomeDude wrote:
>> On Tuesday, 11 December 2012 at 04:36:36 UTC, Walter Bright wrote:
>>> On 12/10/2012 8:21 PM, Walter Bright wrote:
>>>> Yes I understand that, and it is being extensively and heavily used since the
>>>> day I posted it, and no problems have arisen.
>>>
>>>
>>> I should also add that the design was based on extensive discussions about it here last summer.
>>
>> Discussions are not enough. Issues and usage patterns only appear after several months of experience by several users. We need several months of experience before actually gaining any insight and feedback on such a feature.
>
> The root problem is that we have one branch for the unstable, testing, and stable code, and there's no possible way to make anyone happy with only that in place.
>
> The chaotic clash between stableness and unstableness will go away once we have a well thought out development and release process put in place, and it should make everyone happy because it can allow for breaking new changes to be introduced into an unstable branch without messing up the next pre-release version, and updates to pre-release won't mess up the current stable branch because it only gets updated with bug fixes.

I think we all agree on this. Or else we should.

On 12/12/2012 04:45 PM, Iain Buclaw wrote:
>
> Though one of the downsides would be that if I were to leave, so would the site.

For the stability of the project, D needs more commodity-based services like Amazon S3, and less volunteer-hosted, ad hoc services administered by people in their spare time. If D can raise $30k for a one-time conference, raising the money to host project files should be a no-brainer.

On 13 December 2012 08:22, Jeff Nowakowski <jeff@dilacero.org> wrote:
> On 12/12/2012 04:45 PM, Iain Buclaw wrote:
>>
>>
>> Though one of the downsides would be that if I were to leave, so would the site.
>
>
> For the stability of the project, D needs more commodity-based services like Amazon S3, and less volunteer-hosted, ad hoc services administered by people in their spare time. If D can raise $30k for a one-time conference, raising the money to host project files should be a no-brainer.

As it is cloud-based in a way you don't need to know what platform it is running on, it can be moved from location to location with little or no downtime.

-- 
Iain Buclaw

*(p < e ? p++ : p) = (c & 0x0f) + '0';

On 12/13/2012 12:22 AM, Jeff Nowakowski wrote:
> On 12/12/2012 04:45 PM, Iain Buclaw wrote:
>>
>> Though one of the downsides would be that if I were to leave, so would the site.
>
> For the stability of the project, D needs more commodity-based services like
> Amazon S3, and less volunteer-hosted, ad hoc services administered by people in
> their spare time. If D can raise $30k for a one-time conference, raising the
> money to host project files should be a no-brainer.

I thought we had that with github, but then they disabled downloads.

On 2012-12-13 10:01, Walter Bright wrote:

> I thought we had that with github, but then they disabled downloads.

Yeah, we _had_, they just removed it:

https://github.com/blog/1302-goodbye-uploads

-- 
/Jacob Carlborg

On Wednesday, 12 December 2012 at 20:55:52 UTC, Brad Roberts wrote:
> Generous offer.  I've been meaning to build packaging into the auto-tester
> for both release builds and more frequent (nightly or maybe even every
> cycle) builds.  I was going to toss them into s3 with a cloudfront
> distribution in front of that.  Sounds like that needs to go higher on the
> todo list given github's recent changes.
>
> The advantage of CloudFront is fast access all over the planet, though it
> has a cost associated with it.  I'll bear that until it starts to hurt.

+1. If we can get the files hosted on CloudFront, this would be more or less the optimal case – much, much easier than building our own »CDN« out of a few community-provided mirrors, and probably faster for most users as well.

David

On Monday, 10 December 2012 at 00:34:33 UTC, Walter Bright wrote:
> It's time to do a release; to that end we should be working on tidying up the regressions.
>
> This will be the last official D1 release.

Two things which I think we *must* address before the release, otherwise they will hurt us in the long run:


 - https://github.com/D-Programming-Language/dmd/pull/1287 (Make deprecations as warnings the default): Recently, an option to show deprecations as informational messages (i.e. not halting compilation) was added to DMD. Many people, including me, think that this should be the default behavior, but Walter seems to be against it for pretty much unspecified reasons. If we do not finish discussion on this before the release, but then decided to change the default behavior in the future, this will cause quite a bit more confusion as the "-di" switch will already have been released then.


 - Regarding UDAs: Releasing them in the current, not very well-tested state might not be the best idea, because we are effectively setting the design the stone, at least if we don't put a large »experimental« tag on them. However, this is not what I'm talking about here.

Walter's current plan is to also keep the superseded bracket syntax around, making use of it only a warning: https://github.com/D-Programming-Language/dmd/commit/a04cf864b932061ad7b72e7cad8b16fabc6a825a

This is a very, *very* questionable decision, as people won't even know that the syntax is deprecated if compiling without any flags, it adds a big maintenance burden (now the feature must go through the usual deprecation cycle), and there is no reason for keeping it around in the first place. No, backwards compatibility to an unreleased state of Git master is not a valid argument, especially if the feature suddenly appeared without prior notice.


David

On 13 December 2012 16:57, David Nadlinger <see@klickverbot.at> wrote:
> On Monday, 10 December 2012 at 00:34:33 UTC, Walter Bright wrote:
>>
>> It's time to do a release; to that end we should be working on tidying up the regressions.
>>
>> This will be the last official D1 release.
>
>
> Two things which I think we *must* address before the release, otherwise they will hurt us in the long run:
>
>
>  - https://github.com/D-Programming-Language/dmd/pull/1287 (Make
> deprecations as warnings the default): Recently, an option to show
> deprecations as informational messages (i.e. not halting compilation) was
> added to DMD. Many people, including me, think that this should be the
> default behavior, but Walter seems to be against it for pretty much
> unspecified reasons. If we do not finish discussion on this before the
> release, but then decided to change the default behavior in the future, this
> will cause quite a bit more confusion as the "-di" switch will already have
> been released then.
>
>
>  - Regarding UDAs: Releasing them in the current, not very well-tested state
> might not be the best idea, because we are effectively setting the design
> the stone, at least if we don't put a large »experimental« tag on them.
> However, this is not what I'm talking about here.
>
> Walter's current plan is to also keep the superseded bracket syntax around,
> making use of it only a warning:
> https://github.com/D-Programming-Language/dmd/commit/a04cf864b932061ad7b72e7cad8b16fabc6a825a
>

I am confused at this commit also.

-- 
Iain Buclaw

*(p < e ? p++ : p) = (c & 0x0f) + '0';