@tknerr metadata deps not honored #717
Conversation
| @@ -436,3 +436,49 @@ Feature: install cookbooks from a Berksfile | |||
| """ | |||
| Unknown site shortname 'somethingabsurd' - supported shortnames are: | |||
| """ | |||
|
|
|||
| @focus | |||
There was a problem hiding this comment.
I left them in until we can figure out how to fix them (i.e. this PR is incomplete)
|
@sethvargo it should be, you might want to re-write these tests to work with the Berks API server stuff that I introduced and check it out. |
| Installing bacon (0.2.0) | ||
| """ | ||
|
|
||
| @focus |
There was a problem hiding this comment.
@reset can you take a look at this? It's still failing against master. It looks like we still trust local versions...
|
per triage, @andrewGarson to take a look |
ghost
assigned 
|
This looks like a different issue than we have seen in the past. The old resolver would resolve each dependency in order, and if you happened to grab a version of some cookbook that didn't work with a future dependency it didn't care. That's why reordering your dependencies worked. Reordering these dependencies in the test doesn't 'fix' the problem. Something more complex is happening. It looks like some dependency is not being respected at all or a version compare is going awry. Still investigating. |
based upon the version_constraint even though we have already resolved the graph and know what the appropriate version is. This was causing it to grab versions that were too new. Bump ridley version to match master so that bundler will be happy
|
@sethvargo @reset @ivey please take a look at this #cached_cookbook was satisfying based upon the version_constraint instead of the locked_version of a dependency. This was causing berkshelf to resolve the dependencies properly, but then lookup the wrong cached cookbook. |
|
@andrewGarson holy fuck shit. Great find. I don't know how many hours I spent tracing code. We should consider back-porting this and releasing our last 2.0 release. It looks like the build is failing on |
Yeah! 👍 :-) |
|
I can't recreate that failing test locally for some reason and there is no obvious reason for the stack to be blown that way |
|
Just found the problem causing the failure, will work on a fix. |
|
@andrewGarson this looks good, 👍 soon as you get that fix in for the failing tests. |
in #locked_version and #cached_cookbook. This is removed. Instead of getting the cached_cookbook to find the version, just read locked_version
|
@reset @sethvargo you should take a look at this again before merging. |
| @@ -407,3 +407,56 @@ Feature: install cookbooks from a Berksfile | |||
| An error occurred during Git execution: | |||
| """ | |||
| And the exit status should be "GitError" | |||
|
|
|||
| @focus | |||
|
@andrewGarson cool stuff, thanks! 👍 Btw: will this be backported to 2-0-stable or only fixed in master / 3.0? |
|
👍 after removing the @Focus tags. Good catch on the infinite recursion 😄 |
@tknerr metadata deps not honored
|
@tknerr this cannot be easily backported, as the fix is not the same as what @andrewGarson came up with (since the resolver has changed so much) |
|
@sethvargo ok, no problem, just let me know when the next 3.0 pre version is out |
Originally #712, but updated to conform to our test structure. Merge this instead of #712, but let's keep the conversation there. (I think merging this will automatically close #712).
Since this is a branch on RiotGames/berkshelf, anyone on the team can push and make changes.