QGIS » QGIS Application

studing the issue

this issue would affect any provider due to:

https://github.com/qgis/QGIS/blob/release-2_18/src/app/qgisapp.cpp#L7177

delete is done before add feature and there is no way to specify a transaction (btw not all providers support transactions)

my fault, if provider support transaction it can be managed with QgsTransaction

the error is generated here:
https://qgis.org/api/2.18/qgsvectorlayereditbuffer_8cpp_source.html#l00520
e.g. the the update process is in two phases:
1) remove features to merge
2) add the new merged feature
because of https://qgis.org/api/2.18/qgsvectorlayereditbuffer_8cpp_source.html#l00520
and as specified in documentation, if some of the composed steps fail (delete success, and add fail), then we can recover only the last one (the add) during rollback because editingBuffer has already removed from the rollback cache the first step (delete)

in this use case no use og QgsTransaction or QgsTransactionGroup is used

asked in dev list what would be a better solution... especially asked to MarcoH about his commit

reverting (solution 1) does not generate more test regressions (btw coverage in this area is poor)
tested and solution 1 fix the problem. Waiting for community comments to pack PR if solution 1 is accepted

the issue affect qgis3 too

solution 1 works for 3.x

without any answer in list... I'll implement solution 1). Preparing PR

Luigi Pirelli wrote:

Applied in changeset qgis|87116abd72ac52be042c3bff195a3cc931b035e6.

fix applied also to 3.0 e850c82cb01b0ae66d7d3c1f7781fd677ba2ec1f

The same behavior in dev version (revision 313ec55) as described in #16974( using the same attached test data). Difference only in error message:

Errors: SUCCESS: 4 feature(s) deleted.
  ERROR: 2 feature(s) not added.

  Provider errors:
      OGR error creating feature -2: failed to prepare SQL: INSERT INTO bug_examples ( "geom", "name") VALUES (?, ?)
      OGR error creating feature -3: failed to prepare SQL: INSERT INTO bug_examples ( "geom", "name") VALUES (?, ?)

I've to take it again... generally speaking there are two step editing

step1... all edits are in vector editbuffer, thyat does not have any relation with the real data provider. But there are checks about provider capabilities.
step2... changes are saved => the data provider start to play a role

Giovanni Manghi wrote:

This is not completely fixed yet.

The good new is that it seems that the data loss is fixed BUT when trying to merge two geometries in a layer that does not suppurt MULTI geometries do NOT immediately stop the operation as it happens in 2.14. There a dialog shows with the message "Merge cancelled: The union operation would result in a geometry type that is not compatible with the current layer.".

Now the user is not warned, the operation seemingly is ok, but then on save edits QGIS throws an error. If the user meanwhile did more legit digitizing operations he/she has to discard everything, possibly losing a lot of work.

your request is to remove this last action from the editbuffer?

regards

your request is to remove this last action from the editbuffer?

clarified in a chat session.

https://github.com/qgis/QGIS/pull/5223 seems to have addressed the issue. Thanks!