[dataimport] stop ignoring errors on flush That can only result in database corruption.

[dataimport] don't commit on flush Changeset 26cdfc6dd6f8 introduced this confusion for no apparent reason, and it doesn't make much sense.

[test] Add test for dataimport's RQLObjectStore

[dataimport] Stop swallowing errors from commit/flush Silent DB corruption is not OK. Also drop comment from ObjectStore.commit that I don't understand.

[dataimport] Update RQLObjectStore to Connection API Take a connection instead of a session. Don't play games with set_cnxset.

[dataimport] remove _rql heresy 1) ObjectStore knows nothing about rql, it just drops it on the floor 2) RQLObjectStore has a session, it can run rql that way 3) setting an object's "_rql" attribute is not a reasonable user-facing interface Also drop _commit attribute from base ObjectStore, it doesn't use it.

[dataimport] remove dead code The only caller of ObjectStore._put is ObjectStore.create_entity, which RQLObjectStore overrides (and doesn't call up).

[dataimport] do not use sys.exit() to raise missing argument error Related to #3845572

[dataimport] _create_copyfrom_buffer: add the tests Related to #3845572

[dataimport] _create_copyfrom_buffer: do not ignore columns if data is a list Related to #3845572

[dataimport] _create_copyfrom_buffer: fix datetime converter + add time converter Include second and microsecond in the output instead of dropping them on the floor. The time zone is not supported for now, though. Related to #3845572

[dataimport] _create_copyfrom_buffer: raise ValueError if conversion cannot be performed If the conversion fails, there is no reason to continue execution. One should notify the error. Continuing after a misconverted data could corrupt the database. Related to #3845572

[dataimport] _create_copyfrom_buffer: fix separator check in string converter The empty string is a valid replace_sep. Related to #3845572

[dataimport] _create_copyfrom_buffer: put converters into a list Cleans up the code to avoid a succession of ifs. Related to #3845572

merge 3.19.3 into 3.20 branch

Added tag cubicweb-version-3.19.3, cubicweb-debian-version-3.19.3-1, cubicweb-centos-version-3.19.3-1 for changeset 37f7c60f89f1

[pkg] prepare 3.19.3

merge from 3.18 branch

merge 3.17.16 into 3.18 branch

Remove uses of logilab.common.compat.{all,any} They're just aliases to the builtin ones on python 2.5+. If anyone needs convincing: >>> from logilab.common import compat >>> compat.any <built-in function any> >>> compat.all <built-in function all>

remove references to global environment variable APYCOT_ROOT

Add warning messages when enabling remote pickle-based repository access Warn when starting a pyro or zmq-only repo, or when one of these access methods is enabled. Closes #2919295

[web] Fix expiry of anonymous sessions (closes #4154479) Things like: try: foo except: bar else: baz doesn't work very well if you expect baz to run even in the exception case. Plus, session.cnx.check() is a dbapi remnant, so drop it and just look at session.mtime.

[hooks/syncschema] avoid altering a dropped table If the constraint's rdef is being removed, CWConstraintDelOp doesn't need to do anything. Otherwise it may try to alter a removed table/column. Closes #4154549

[js] fix name error

[devtools] improve error message when postgresql tools are missing By default in at least Debian, some pg tools are not present in the PATH. Or they may not be installed. But the tests tools expects them to be in the PATH, and give an unhelpful 'No such file or directory' backtrace if they're not found. To help devs using the pg tests we improve the error message.

[server] Replace non portable strftime formatter (closes #4132161)

[web/request] use a picklable Counter object for tab index counters Related to #1381328.

[web/auth] The auth retriever's authenticated() callback takes a session, not a cnx We don't have a cnx at that point.

[test] Fix the query-log-file test Testing this functionality in a CubicWebTC is awkward because the wrapping happens in handle_request, which tests basically bypass (they call core_handle directly, and do their own magic for linking the request with a cnx). The test method worked when ran on its own, but not together with the rest of the test class. So use a CubicWebServerTC instead, which goes through the whole stack.