Trying to fix some of the HHVM unittests #12336
Conversation
|
Seems like this indeed "fixes" that error, however throwing a fatal error because of this: facebook/hhvm#7320 Seems as if we would only support the very latest version of HHVM with Postgres. Can we have such a requirement or do we have to introduce a conditional for that function call? |
|
So... with this PR, the errors would be reduced from 80 to 11. Is it worth that? |
|
Yes it's worth |
|
can't this go for staging and 3.7.x too? |
|
Well, I tried changing the base branch but the PR would've drug in the entirety of the 4.0 branch. But yes, this should go into staging instead of 4.0 only. |
|
We can merge it against 4.0-dev and we can open a new PR against staging 😄 Done: #12341 |
|
These are exactly the changes I have made in my test branches pending release of hhvm 3.15.2 or 3.16 which will resolve these issues. facebook/hhvm#7399 (comment) |
|
BTW some of these issues were cataloged a while back in #10220 although the connection issue was new in 3.15 with the inclusion of pgsl.so in HHVM as noted in facebook/hhvm#7399 (comment) |
|
@Hackwar According to the documentation spaces or semi-colons should work,
Fortunately HHVM is working on a fix. But I do agree this is a valid fix/work around (personally, I think in most cases spaces are the worst choice for delimiters) I have been hesitant to submit the change for our call to looks good on code review. |
|
So with those comments, we could merge this. 😄 |
|
The change to our call of |
|
The HHVM patch to fix the DSN separators to allow semicolon or space has been merged, the HHVM 3.16.0 STS release will have this, and is on the list of potential cherry picks for HHVM 3.15.3 LTS. HHVM 3.16.0 STS is expected around 10/24/2016 and will automatically be the version tested on Travis CI since we are testing against HHVM latest. |
|
so all we have to do is wait? |
|
@andrepereiradasilva There is no harm in merging as it's a semantics change to our unit test, but yes it will also be fixed by just waiting. I would suggest backing out the |
|
Depends what we consider a minimum supported HHVM version to be. If it's something older than 3.15.2 it'll be needed. |
|
@mbabker Since hhvm 3.15.0 is the first version with pgsql.so in core I would suggest that would be the earlist LTS version to consider. 3.15.2 is also the first version were tests do not error out on memory. As noted in #10220 (comment) we can pick up any recent LTS version of HHVM for testing on Travis CI or latest or nightly. we can also test with and without HHVM's php 7 mode (although the HHVM php 7 mode is currently broken due to facebook/hhvm#7198 ) We still need a fix for facebook/hhvm#2060 and to solve the JInstallerAdapterTest failures or determine if the JInstallerAdapterTest failures are an HHVM bug before we can determine a minimum HHVM version to support. I've been pushing HHVM to include the important bug fixes we need in the patches of 3.15 so that hopefully 3.15 can be a valid LTS release to target for a supported version. |
|
Does the CMS actually work on HHVM practically (and by work I mean is the core extensions + functionality useable not just do the unit tests pass) |
|
Who actually runs HHVM outside of Travis to test that? 😆 |
|
It's a lot of work being done if no one uses hhvm
|
|
if anyone has a full siteground account, I believe that host does HHVM. I've seen a few comments from people who tried running J3.4 on HHVM but ran into a lot of issues. Some of the past items have been fixed in the CMS or in HHVM. At the moment the failures in the unit tests suggest to me that it's unlikely for the CMS to actually run on HHVM without issues. But I haven't tried running anything other than unit tests. |
|
Only on their Cloud Hosting plans - not the regular ones On 18 October 2016 at 15:12, Walt Sorensen [email protected] wrote:
Brian Teeman |
|
and it is only an option not a default On 18 October 2016 at 15:32, Brian Teeman [email protected] wrote:
Brian Teeman |
|
Best priced options for HHVM that I could find on a quick search was Google cloud at $5/mo or free trial and A2 Hosting at $5/mo red hat openshift claims to have a free HHVM option in their developer preview, but I couldn't get it to launch into my account. Heroku says it offers HHVM as a "highly experimental option", but I haven't had time to deal with a platform that relies heavily on command line for setup and activation. I was looking at heroku as an option since heroku has a free option that can be sandboxed for testing. |
|
The question is "are there any users with hhvm" On 18 October 2016 at 16:45, Walt Sorensen [email protected] wrote:
Brian Teeman |
As I said, I've seen a few comments from people who tried running J3.4.6 on HHVM but ran into a lot of issues. I haven't seen anything more recent. |
|
We still need the DSN separator portion of this backported to staging since #12341 was closed. (of course, it won't be needed whenever HHVM 3.16.0 gets released in the next few weeks or hopefully 3.15.3 if they merge the fix for the release in the next few weeks). |
According to the documentation, a DSN connection string should use semicolons as delimiter and not spaces. It should also not make any difference for the normal PHP implementation. Since we have currently ~80 errors in the test suite when running on HHVM and most of these are related to this DSN string, I'm trying to fix this. Funnily, I don't have access to HHVM, so I'm going to try to fix this by running this through the unittests again and again... 🙊 😄