Closed (fixed)
Project:
Project
Version:
6.x-1.x-dev
Component:
Releases
Priority:
Critical
Category:
Bug report
Assigned:
Unassigned
Reporter:
Created:
10 Jul 2010 at 14:54 UTC
Updated:
28 Jul 2010 at 21:40 UTC
Jump to comment: Most recent file
Comments
Comment #1
perusio commentedSame here. Drush reports a MD5 checksum error though in reality is a 404 issue. The page cannot be found. If I download the recommended release instead of dev everything works ok.
Comment #2
roball commentedComment #3
avpadernoI can confirm the issue. It's not a problem with the FTP server, or DNS because it only happens with the development snapshots, not with the official releases.
I tried to download CCK 6.x-2.7, and I was able to download it; when I tried to download the development snapshot for Drupal 6, I got an error 404 page. The page was not from Drupal, though.
Comment #4
drummThere were some project* and solr changes pushed live last night. Attached are the project changes. I'll look around for relevant logs.
Comment #5
drummWhitespace ignored
Comment #6
drummWith whitespace ignored.
Comment #7
damien tournoud commentedHm. D'oh.
The current packaging script will call
cleanup_failed_build()as soon as the build failed, andcleanup_failed_build()returns FALSE when a-devbuild doesn't have to be regenerated.Comment #9
hunmonk commentedthere was a cleanup_failed_build() function added to the packaging script, which is supposed to remove tarballs in the event that the build fails. it tested out fine on scratch, and it's pretty simple logic, but it's the only thing i see in here that could be causing the problem.
Comment #10
Poieo commentedSubscribing...
Comment #11
damien tournoud commentedI committed the following in the drupal.org repository as a quick-fix. Reassigning to Project to discuss an eventual better fix.
Comment #12
dwwYes, #777068: profile-only release tarballs not cleaned up if other tarballs fail to build was done wrong. It was causing the -dev tarballs to churn -- every other packaging run deleted the "failed" build. ;)
The logic was supposed to only happen on profile packages, but it's happening on all contribs.
Furthermore, the logic is wrong. DamZ's quick fix is sort of the right approach, but I think it might create other problems. I think we're going to need more than true/false return values here...
Comment #13
Poieo commentedSorry to dirty up an issue that is way over my head, but does this mean the missing dev's will be available again at some point tonight. Or, will the maintainers need to re-upload their dev releases?
Comment #14
dww@Poieo: The missing -dev are being rebuilt as I write this. It's up to the modules beginning with the letter "o". ;) No, you don't have to do anything (and any attempt to "fix" this by module maintainers would almost certainly make things worse).
Re the return values -- the script really needs to know the difference between "i did useful work that needs to be updated into the release node", "I didn't do anything", and "I hit errors". So, instead of trying to return TRUE/FALSE, we should really just return 'success', 'no-op', and 'error'. Then, the code is fully self-documenting and we can properly handle all 3 cases correctly.
Comment #15
dww(Better title to reflect the actual nature of this bug)
Comment #16
hunmonk commentedattached patch should fix the problem in a more robust fashion, using the logic three string values dww proposed in #14
Comment #17
roball commentedNo where the dev tarballs are being rebuilt, they *all* are dated 2010-Jul-10 even if they are in fact very old, causing the update module on Drupal installations pinging for "new releases available" for all modules and themes that you have running at a dev version...
Comment #18
dwwroball: nothing we can do about that...
Comment #19
dwwPatch #16 looks exactly like what I was proposing. I haven't tested it, but visually that's the right fix.
Comment #20
a5342346 commentedsubscribe
Comment #21
hunmonk commentedtested the patch in #16 thoroughly in a local environment, committed to 6.x-1.x-dev, deployed on drupal.org
Comment #22
a5342346 commentedsound likes "it's fixed"
can you pls comment as to what end-users should now see/do?
after
drush cc all
drush ref
@ drush up/dl i'm still getting " ... is corrupt (wrong md5 checksum)" errors.
still need to wait for something to propagate?
Comment #23
merlinofchaos commentedCan dev tarballs be restored from backup? That filedate is actually kind of important information. I guess doing so is probably more pain than it is worth, but...ugh. I know I personally like to compare -dev filedates to release filedates to guess if there's been recent activity, and now that information could be wrong. Vastly wrong in some cases.
Comment #24
merlinofchaos commentedHmm. CTools (which begins with C) does not have the -dev tarball restored. My reading is that they should all be regenerated by now?
Comment #25
jlmeredithI am experiencing the tarballs missing for views_bulk_operations and uc_views
Comment #26
roball commentedMe too - there are still missing dev tarballs, eg. http://ftp.drupal.org/files/projects/at_koda-6.x-1.x-dev.tar.gz (#850394: at_koda-6.x-1.x-dev.tar.gz download link 404). Those projects are affected really critical since they do not have an official release yet and have thus no tarball to download from the project site at all!
Comment #27
dwwhunmonk deployed the fix, but it doesn't look like the every 12-hours packaging run happened yet.
I'm running the script manually again now. Once that finishes, I'll run it again to make sure the bug is actually gone. ;)
Stay tuned...
Comment #28
dww"Done packaging releases from branches: 9258 built, 9585 considered."
The rsync to ftp.drupal.org should be done by now.
All the release history XML files are being regenerated now. It's taking a long time, since this is the worst case -- basically it has to regenerate the history for all 5000 contribs. :( It's 1:36am here, so I need to sleep. I'll take a look at this first thing in the morning and see if the release history is cool, and make sure that the tarballs aren't needlessly rebuilt when the script runs again.
Comment #29
jlmeredithThanks to those that stayed late to work this out! Your fellow Drupal peeps appreciate what you do!
Comment #30
jlmeredithDrush is reporting an invalid checksum when issuing the command -drush dl views_bulk_operations-6.x-3.x-dev.tar.gz
Comment #31
AntiNSA commentedsubscribe
Comment #32
AntiNSA commentedhttp://drupal.org/node/850786
I hate posting this in the general forum instead of issue ques because I cant see the responses when I click on "My issues"
However, I awoke as I normally do to find out that the "Available Updates" was really crazy today. I update everyday, but today found 40 of my modules had available updates on the same day? Thats crazy.
When I downloaded them many were only 1k big, and gave rar errors.
Any ideas?
Comment #33
AntiNSA commentedYou know, with something so huge as this dont you think someone should post this info on the front page of drupal so everyone knows whats up?
Why even have to ask about it? It seems like a pretty important issue. Would have saved me a lot of time from downloading 40 modules.
Comment #34
yesct commented#850724: http://ftp.drupal.org/files/projects/location-6.x-3.x-dev.tar.gz not found is a duplicate of this issue. ... and is still a problem, but #28 looks like it might resolve itself in time.
Comment #35
dww@All: Stuff happens. Please relax. If you use -dev modules, it's the Wild West, and crazy things can happen. That's one of many reasons why it's so discouraged to rely on -dev. No, I don't think this bug is worthy of front page news. -dev is fragile, and unstable. Get over it. ;)
The http://ftp.drupal.org/files/projects/location-6.x-3.x-dev.tar.gz tarball is working fine for me.
I just re-ran the packaging script and it did exactly what it's supposed to. It ignored most -dev, and the ones it rebuilt I did some spot checking and the maintainer had in fact committed some changes.
Yes, it's a hassle that all the tarball datestamps are wrong. To that end, we're considering restoring them all from backup, then re-running packaging to re-generate anything that happened since the backup. However, that's going to take a while. And, while it's happening, update status is going to be really confused, too. If we do it, it'll be later this afternoon (after the World Cup Final!). ;) Right now, a full backup is running, so that's going to take probably over an hour itself.
Comment #36
roball commentedWhy not changing the drupal.org packaging script to set the value of datestamp (into the .info files) to the date of the last CVS commit instead of the current time?
Comment #37
dww@roball: Because that'd be wrong. The file metadata should match the files themselves. Plus, the .info files are only a tiny part of the problem here. We also need the datestamp in the {files} table for these files, so that things are propagated to the release history XML files, etc, etc.
Comment #38
roball commentedThe script could also set the file's last-modified timestamp to that date.
Comment #39
dwwI should have reiterated: "I have better things to do than worry about how -dev behaves in update status". ;) It's a mess. -dev sucks. Get over it.
Comment #40
AntiNSA commentedJust like nearly 2 years using drupal for me and dev fixes a lot of my problems all the time so its always first in my workflow.... Scary to see the hearbeat of drupal quiver.... ok as long as it works again. I dont care anything about football . Drupal is much better.
Comment #41
AntiNSA commentedok I still have a problem using --- reports ---- available updates..... I can still see every dev with an update... any idea?
Comment #42
Jeff Burnz commentedI'm still getting the error with my D6 -devs, D7 -devs are ok, should I commit a small change to trigger a rebuild for the D6's, fyi these all start with "a".
Comment #43
dww@Jeff Burnz: Please provide specific links that don't work -- I don't have time to go hunting and guessing which links you're trying. ;)
@All: nnewton was worried about the restore from backup, and given that it'd require a bunch of extra work to make that all possible (it's not as simple as re-running the packaging script, after all), we aborted the plan to restore the old -dev tarballs. So yeah, if you have a lot of -dev, update status is going to be telling you to update.
Comment #44
roball commentedJeff - another workaround would be to just release an alpha version, wouldn't it?
Comment #45
AntiNSA commented#43
Every dev version I have says there is an update available, even some releases recommending dev over stable when I click on reports--available updates... i think that is what he means. When downloading these 1k rar files they are not truly updates.
So I cant get any updates for drupal for 2 days.
Comment #46
avpadernoI have marked #851298: Available update report, all dev versions dated 2010-Jul-11 as duplicate of this report.
Comment #47
AntiNSA commentedCAn you tell me is the dev system for upgrading modules now pernamently dead? What is the suggested new way to get module updates?
Comment #48
sunfish62 commentedIn re #35: I understand your points about development packages, but disagree about your conclusion about notifying the Drupal community. I think *some* kind of note or announcement in a prominent place should be made.
Like it or not, there are MANY websites that use development modules that are affected by this, and to bury this in a forum is disrespectful of all the people who manage those sites.
Comment #49
AntiNSA commentedabsolutely.. dev has been recommended under some updated modules even...
Comment #50
dwwRe #47: No, it's not permanently dead. If you "upgrade", then it'll be working again if new -devs are packaged from here on out. Just pretend like every module you're using actually had a commit on July 11th, which is certainly possible.
Comment #51
gerhard killesreiter commentedPeople who think it is a good idea to use dev-releases on live sites will now hopefully know better why this is not recommended.
There's nothing to do here.
Comment #52
AntiNSA commented#51 you are quite wrong. There are several modules were dev releases are recommended. This is not an individual module issue. This is a drupal project dostrobution issue.
#50 So you are saying, if there is a new dev released for all of these modules which are listed with july 11th dates, and the release is newer than July 11th, we will no longer see July 11th as the updated date?
Comment #53
dww@AntiNSA: As I said, just pretend every module on drupal.org had a commit on July 11th. So, all the tarballs were rebuilt. Everything works like it always did. If someone commits again on July 12th, the tarball is rebuilt and update status will tell you that the July 12th copy is the most recent.
Comment #54
AntiNSA commentedok thanks I now understand.
Comment #55
michelle@Gerhard: In an ideal world, yes. Unfortunately, there are far too many maintainers who put up interesting and useful modules on drupal.org and then let them languish in development without so much as an alpha forever. Or fix bugs in the dev of a stable release and then go years without another release. So the fact is that many people do use dev versions not because they are stupid but because poor maintenance doesn't leave them much choice.
The flip side of that is that there are some dev snapshots that are very good. I often run the dev snapshots of cck/views/panels/ctools because they are rarely broken. I take care, of course, when upgrading from dev to dev because even they make mistakes now and then but, on the whole, they are quite usable.
Michelle
Comment #56
sunfish62 commented@Gerhard: Michelle has stated this more eloquently than I could have. +1 Michelle
David
Comment #57
Junro commentedsubscribe,
We can deal with this problem until a fix. It's a little confusing but it's easy to know what the real new dev version of modules are.
Meanwhile, It's very annoying for new drupal users that built their website from start.
Comment #58
pasqualleno, the dev release can't be recommended, it is not possible..
Michelle: The dev release could be quite usable, but the stable release is more stable than development release, by definition.
If you want a stable release, then create an issue in that module's issue queue..
this issue is fixed.
Comment #59
Junro commented@Pasqualle
I'm not agree with you at all.
You CAN'T make a complete website with only recommended version. It's impossible. You have to use dev. version.
As Michelle said
Poor maintenance, it's not a choice for some modules. The problem of compatibility between modules is inevitable. Maintainers can't always make a recommended release after each bug.
It's something we had learned to deal with.
Comment #60
a5342346 commented> no, the dev release can't be recommended, it is not possible..
well ...
drush pm-releases `drush pm-list | cut -d "(" -f 2 | cut -d ")" -f 1` | grep -i recom | grep dev
varnish 6.x-1.x-dev 2010-Jul-11 Supported, Recommended, Development, Installed
contact 6.x-2.x-dev 2010-Jul-11 Supported, Recommended, Development
devel 6.x-1.20 2010-Apr-21 Supported, Recommended
devel_themer 6.x-1.x-dev 2010-Jul-11 Supported, Recommended, Development, Installed
expire 6.x-1.x-dev 2010-Jul-11 Supported, Recommended, Development, Installed
dialog 6.x-1.x-dev 2010-Jul-11 Supported, Recommended, Development, Installed
Comment #61
AntiNSA commented#60 theres more too :) Its all good though... everything is working again. let it go guys.. this cant be the thread about evil dev's :)
Comment #62
sunfish62 commented@ 61: I'd let this drop, too--Gladly. I just wish that open lines of communication could be encouraged in the Drupal community, instead of finger pointing.
It appears to me like someone committed code prematurely (with substantial impact across the Drupal community), and then, instead of owning to that fact openly, they hoped that the problem would go away. So, instead of an announcement placed prominently on the site advising the community of users that an error was made to the core Drupal website and that the inconvenience is unavoidable, we get a thread under the Drupal issue queue in which the user community is harangued and made to seem like it is the problem (cf. #35, 39 and 51).
Comment #63
perusio commentedI agree. It's not about pointing fingers but about solving the issue instead of blaming someone because "he/she doesn't know better" than use dev versions.
Owning the problem is the first step towards solving it.
Comment #64
damien tournoud commentedThe problem *has been* fixed, and a decision has been reached. This thread is now locked.
[edited to increase politeness]