| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
the default connection permissions for those who don't have a predefined (or therefore have a "custom") permissions role. Unfortunately this includes most people that were using this software more than a month ago. The real changes are that the SELF address book entry no longer holds "auto-permissions" but instead holds your "default permissions" (if you have a pre-defined role, the defaults will be pulled from the role table).
The auto permissions have moved to a pconfig (uid.system.autoperms). A DB update will move these settings into their new homes.
What used to be the "Auto-permissions settings" page is now the "default permissions settings" page and a checkbox therein decides whether or not to apply the permissions automatically. A link to this page will only be shown when you have the "custom" role selected.
With luck nobody will notice anything wrong. But at least for the next few days, please review permissions that have been assigned to new connections (either automatically or manually) and make sure they make sense (e.g. they aren't "nothing"). You still need to take action when seeing a message "permissions have changed but not yet submitted" as we always let you review and perhaps adjust the settings _before_ a connection is established (unless you have autoperms turned on).
|
| | |
|
| |
| |
| |
| | |
this over and tested a lot of edge cases, and thought about from every angle I can think of to prevent looping. I don't *think* this can loop. I also doubt that this is the problem at friendicared.de, but I don't know for sure what that problem might be.
|
| |
| |
| |
| | |
back on
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There were 11 main types of changes:
- UPDATE's and DELETE's sometimes had LIMIT 1 at the end of them. This is not only non-compliant but
it would certainly not do what whoever wrote it thought it would. It is likely this mistake was just
copied from Friendica. All of these instances, the LIMIT 1 was simply removed.
- Bitwise operations (and even some non-zero int checks) erroneously rely on MySQL implicit
integer-boolean conversion in the WHERE clauses. This is non-compliant (and bad programming practice
to boot). Proper explicit boolean conversions were added. New queries should use proper conventions.
- MySQL has a different operator for bitwise XOR than postgres. Rather than add yet another dba_
func, I converted them to "& ~" ("AND NOT") when turning off, and "|" ("OR") when turning on. There
were no true toggles (XOR). New queries should refrain from using XOR when not necessary.
- There are several fields which the schema has marked as NOT NULL, but the inserts don't specify
them. The reason this works is because mysql totally ignores the constraint and adds an empty text
default automatically. Again, non-compliant, obviously. In these cases a default of empty text was
added.
- Several statements rely on a non-standard MySQL feature
(http://dev.mysql.com/doc/refman/5.5/en/group-by-handling.html). These queries can all be rewritten
to be standards compliant. Interestingly enough, the newly rewritten standards compliant queries run
a zillion times faster, even on MySQL.
- A couple of function/operator name translations were needed (RAND/RANDOM, GROUP_CONCAT/STRING_AGG,
UTC_NOW, REGEXP/~, ^/#) -- assist functions added in the dba_
- INTERVALs: postgres requires quotes around the value, mysql requires that there are not quotes
around the value -- assist functions added in the dba_
- NULL_DATE's -- Postgres does not allow the invalid date '0000-00-00 00:00:00' (there is no such
thing as year 0 or month 0 or day 0). We use '0001-01-01 00:00:00' for postgres. Conversions are
handled in Zot/item packets automagically by quoting all dates with dbescdate().
- char(##) specifications in the schema creates fields with blank spaces that aren't trimmed in the
code. MySQL apparently treats char(##) as varchar(##), again, non-compliant. Since postgres works
better with text fields anyway, this ball of bugs was simply side-stepped by using 'text' datatype
for all text fields in the postgres schema. varchar was used in a couple of places where it actually
seemed appropriate (size constraint), but without rigorously vetting that all of the PHP code
actually validates data, new bugs might come out from under the rug.
- postgres doesn't store nul bytes and a few other non-printables in text fields, even when quoted.
bytea fields were used when storing binary data (photo.data, attach.data). A new dbescbin() function
was added to handle this transparently.
- postgres does not support LIMIT #,# syntax. All databases support LIMIT # OFFSET # syntax.
Statements were updated to be standard.
These changes require corresponding changes in the coding standards. Please review those before
adding any code going forward.
Still on my TODO list:
- remove quotes from non-reserved identifiers and make reserved identifiers use dba func for quoting
- Rewrite search queries for better results (both MySQL and Postgres)
|
| |
|
| |
|
| |
|
|
|
|
| |
prevent them from recursing
|
|
|
|
| |
a missing top-level post to match it with. So we'll send a request back to the sender that you've never seen this thread and please send a fresh copy of the entire conversation to date. We could soon have posts in the matrix from different platforms from days gone by, which have been migrated into the modern world. We'll be polite and not deliver these to everybody. However, if someone comments on one of these antique threads we wouldn't be able to see it in our own matrix because we won't have a copy of the parent post. So this rectifies that situation. Be aware that item deletion may need to change to keep "hard deleted" items indefinitely so that they don't keep coming back. We'll have to null out the important data of the former item to accomplish the deletion aspect.
|
|
|
|
| |
being relayed more than once - as it's a huge drain on resources. But last time I tried this, wall-to-wall comments stopped getting relayed. This checkin should do the right thing in both conditions.
|
|
|
|
| |
directory update packet.
|
|
|
|
| |
is inconsistent data which you think you trust.
|
| |
|
| |
|
| |
|
|
|
|
| |
change but it is unrelated to current issues, basically if no primary was set we were setting everything as primary.
|
|
|
|
| |
issue #633
|
| |
|
|
|
|
| |
delivery chains, which were not adequately accounted for in the earlier checkin.
|
| |
|
|
|
|
|
|
|
|
| |
This checkin implements route matching of comments so that they are only accepted from the same route as the top-level post they are attached to. This way there should be no mis-match of permissions between any posts in the thread. It may not be completely compatible with comments posted in the past (though I've tried to be, there may be some minor issues). In addition it seems that relaying was invoked more often than necessary - especially when a duplicate post arrived which was not processed because the edited time hadn't changed - it still invoked relaying. This fix should improve site performance considerably for comments cross-posted to forums; which got bounced around a bit and delivered redundantly for no reason.
Roll this back *only* if it causes a meltdown or comment loss is "serious" (as in OMG people are dying, make it stop!). If we can get past 24 hours without serious issue we need to get everybody onto this code. There may be some minor comment loss (mostly affecting new comments to older posts or likes of older comments) until the majority of sites have moved to the new code.
It may be difficult or impossible to deliver comments to posts that pre-date the addition of source routes (April 1, 2014) to anybody but the top-level post author at his/her primary hub. We may wish to close comments on these posts, but let's see how we go before doing that.
|
| |
|
| |
|
|
|
|
| |
problems. Also clean up fetch_url/post_url header option
|
|
|
|
| |
("friends") and total_feeds both when importing channels and subsequently when syncing clones. Limits are based on the local system - additional entries are silently dropped.
|
| |
|
|
|
|
|
|
| |
now been split off into its own function. Cross fingers because this is core functionality which was tried and true, well tested; and now it's a bit different.
Please revert this if the matrix breaks in the next few hours.
|
|
|
|
| |
don't rush me. This is going to become a fundamental part of zot. It deserves careful consideration.
|
|
|
|
| |
still a fair ways from being complete and is not ready for prime time. Basically we'll let a channel send out a public message saying "these are my currently approved locations" and anything that isn't in the list will be marked deleted. We'll send out this message when locations change somehow - either through direct personal involvement (hub revoke, change primary, channel import) or during a system rename or "find bad/obsolete hublocs" activity. This way we won't have clones sending back location info we just got rid of and re-importing the bad entries.
|
|
|
|
| |
issues we'll just make it '-'.
|
|
|
|
| |
easily mis-typed sequence '0000-00-00 00:00:00'
|
| |
|
| |
|
| |
|
|
|
|
| |
a whole slew of FIXMEs
|
| |
|
|
|
|
| |
notifier is setup to take hublocs, not xchans.
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
xchan_import - I suspect we don't have anything useful at all.
|
| |
|
|
|
|
| |
public, but they didn't set the private flag.
|
| |
|
| |
|
|
|
|
| |
can later add alternate encodings.
|
|
|
|
| |
across sites.
|