Le dim. 25 avr. 2021 à 14:31, Mark Struberg <strub...@yahoo.de.invalid> a écrit :
> broken before, broken afterward. Sorry that is not a use case which > justifies tons of effort on our side. > Tons as stashing a line? ;) Im fine if we do the 3.1.3 and 3.2.0 at the same time, does not require more efforts if done this way IMHO. We can also investigate the doc (not only jira which is never read by users) option (didnt test if it was configurable in the unit props but i assume so) and highlight it properly on the announcement (like a "danger" banner). But we must focus on users and not us IMHO so we must have a solution to limit properly user breaking changes on used features whatever we like it or not and even when we assume we made it better like here IMHO. Also note that you assume it was broken but tons of apps work well limited to int (vs long). Im fine with both options but im not fine we swallow it silently. > LieGrue, > strub > > > Am 25.04.2021 um 14:08 schrieb Romain Manni-Bucau <rmannibu...@gmail.com > >: > > > > Le dim. 25 avr. 2021 à 13:47, Mark Struberg <strub...@yahoo.de.invalid > <mailto:strub...@yahoo.de.invalid>> a > > écrit : > > > >> No, not a change. > >> > >> If an Entity had a 'double' field it was mapped to NUMERIC. Which has no > >> fraction digits at all > >> > >> 7.23456789d would be stored as 3. > >> So we can rather safely assume that people worked around this by using > >> float, BigDecimal, etc. > >> Thus no change for existing projects. > >> > > > > Sure that some projects use the default and do diff between snapshots and > > previous release ddl based on that so a change ;). > > > > > >> LieGrue, > >> strub > >> > >> > >>> Am 25.04.2021 um 13:39 schrieb Romain Manni-Bucau < > rmannibu...@gmail.com > >>> : > >>> > >>> AFAIK org.apache.openjpa.jdbc.meta.MappingInfo#mergeColumn will use > >>> getJDBCType and in case of HSQLDB it will move from NUMERIC to > BIGINTEGER > >>> which is a breaking change for existing apps - detail in > >>> https://issues.apache.org/jira/browse/OPENJPA-2648 < > >> https://issues.apache.org/jira/browse/OPENJPA-2648>. This kind of > change > >> is > >>> fine for 3.2 but not 3.1.3 we should release IMHO. Same kind of change > >>> happent to SQLServer (for blob IIRC) and a few other dicts IIRC (not > all > >>> impact DDL but it can impact the runtime and for sure impacts the dict > >>> customizations (not sure how important it is but it happens our dicts > are > >>> extended to use some driver specificities). > >>> > >>> I'm not sure of the SQLServer dict usage but HSQLDB one is really used, > >>> even with 1.8 branch so can be neat to at least ensure this part does > not > >>> break *by default* for next immediate release. > >>> > >>> Romain Manni-Bucau > >>> @rmannibucau <https://twitter.com/rmannibucau < > >> https://twitter.com/rmannibucau>> | Blog > >>> <https://rmannibucau.metawerx.net/ <https://rmannibucau.metawerx.net/ > >> > >> | Old Blog > >>> <http://rmannibucau.wordpress.com <http://rmannibucau.wordpress.com/>> > >> | Github <https://github.com/rmannibucau < > https://github.com/rmannibucau>> > >> | > >>> LinkedIn <https://www.linkedin.com/in/rmannibucau < > >> https://www.linkedin.com/in/rmannibucau>> | Book > >>> < > >> > https://www.packtpub.com/application-development/java-ee-8-high-performance > >> < > >> > https://www.packtpub.com/application-development/java-ee-8-high-performance > >>>> > >>> > >>> > >>> Le dim. 25 avr. 2021 à 13:14, Mark Struberg <strub...@yahoo.de.invalid > >> <mailto:strub...@yahoo.de.invalid>> a > >>> écrit : > >>> > >>>> No, the DDL remains the same. We just switched to using java.time > types > >> in > >>>> Oracle natively on the JDBC level. But the type names stood the same. > >>>> > >>>> LieGrue, > >>>> strub > >>>> > >>>> > >>>>> Am 24.04.2021 um 22:41 schrieb Romain Manni-Bucau < > >> rmannibu...@gmail.com > >>>>> : > >>>>> > >>>>> Le sam. 24 avr. 2021 à 22:08, Mark Struberg > <strub...@yahoo.de.invalid > >>>> <mailto:strub...@yahoo.de.invalid <mailto:strub...@yahoo.de.invalid > >>> > >> a > >>>>> écrit : > >>>>> > >>>>>> Romain, I have about 20 BIG apps which I did run past the new > version. > >>>> We > >>>>>> needed to change maybe 3 lines of code over all those apps. It's > >> really > >>>>>> edge cases! And we didn't totally turn around the dictionary at all. > >> We > >>>>>> just improved it. People likely were forced to give names manually > >>>> already > >>>>>> because it would simply not run without in the old version when a > >>>> reserved > >>>>>> word was not catched as invalid column name. > >>>>>> Same goes for the tag handling. If someone did write a query "... > and > >>>> case > >>>>>> .. when... x.bla=true" then it did just NOT work before. So people > >>>> likely > >>>>>> had to use a parameter instead. This still works, but now also the > >>>> variant > >>>>>> with using a tag does work. Again, nothing is broken which did work > >>>>>> before... > >>>>>> > >>>>>> I've deployed 3.2.0-SNAPSHOT now manually. Happy to get feedback if > >>>>>> something is broken. > >>>>>> > >>>>> > >>>>> Some keywords change make the generated ddl different now IIRC. > >>>>> This is the part i had in mind which is a blocker to upgrade since > ddl > >>>> are > >>>>> no more stable. > >>>>> Not all were forced to be changed for ddl generation - depends > >> driver/db > >>>>> version. > >>>>> Dont recall if it was about hsqldb or another one but dont think we > can > >>>>> release it without a 3.1.3. > >>>>> Doing a subdict keeping old one can mitigate it and enable to bypass > >>>> 3.1.3. > >>>>> > >>>>> > >>>>>> LieGrue, > >>>>>> strub > >>>>>> > >>>>>> > >>>>>>> Am 24.04.2021 um 18:32 schrieb Romain Manni-Bucau < > >>>> rmannibu...@gmail.com <mailto:rmannibu...@gmail.com> > >>>>>>> : > >>>>>>> > >>>>>>> Le sam. 24 avr. 2021 à 17:13, Mark Struberg > >> <strub...@yahoo.de.invalid <mailto:strub...@yahoo.de.invalid> > >>>>>> <mailto:strub...@yahoo.de.invalid <mailto:strub...@yahoo.de.invalid > >>> > >> a > >>>>>>> écrit : > >>>>>>> > >>>>>>>> Who needs it? > >>>>>>> > >>>>>>> > >>>>>>> Most 3.1.2 users, we had several expected fixes which shouldnt > >> trigger > >>>>>> more > >>>>>>> than a build revalidation (and in particular not a multidb > >> validation). > >>>>>>> > >>>>>>> > >>>>>>>> For now it's not on my personal agenda as the fixes are good and > >>>> moving > >>>>>> to > >>>>>>>> 3.2.0 is for me personally just a signal to our users that they > >> should > >>>>>> not > >>>>>>>> blindly do an upgrade without a small test run first. The changes > >> are > >>>>>>>> really subtle, but again, they _might_ break in some edge cases. > For > >>>>>> vast > >>>>>>>> majority of people it will be a drop in replacement. > >>>>>>>> > >>>>>>> > >>>>>>> We touched dicts and mapping so i expect it to not be that > >> transparent. > >>>>>>> At least it breaks ddl which is unexpexted > >>>>>>> > >>>>>>> > >>>>>>>> LieGrue, > >>>>>>>> strub > >>>>>>>> > >>>>>>>>> Am 24.04.2021 um 13:16 schrieb Romain Manni-Bucau < > >>>>>> rmannibu...@gmail.com <mailto:rmannibu...@gmail.com> > >>>>>>>>> : > >>>>>>>>> > >>>>>>>>> What's the plan for 3.1.3 release then? > >>>>>>>>> > >>>>>>>>> Le sam. 24 avr. 2021 à 11:38, Mark Struberg > >>>> <strub...@yahoo.de.invalid <mailto:strub...@yahoo.de.invalid> > >>>>>>> > >>>>>>>> a > >>>>>>>>> écrit : > >>>>>>>>> > >>>>>>>>>> I'll move forward and update to 3.2.0-SNAPSHOT > >>>>>>>>>> > >>>>>>>>>> LieGrue, > >>>>>>>>>> strub > >>>>>>>>>> > >>>>>>>>>>> Am 19.04.2021 um 08:35 schrieb Francesco Chicchiriccò < > >>>>>>>>>> ilgro...@apache.org>: > >>>>>>>>>>> > >>>>>>>>>>> On 18/04/21 12:31, Mark Struberg wrote: > >>>>>>>>>>>> Hi folks! > >>>>>>>>>>>> > >>>>>>>>>>>> We fixed a lot of tickets since the last release. Some of them > >>>> also > >>>>>>>>>> change/fix the behaviour slightly. There are a few main tickets > >>>> which > >>>>>> do > >>>>>>>>>> not introduce a big change, but might very subtly break existing > >>>> apps > >>>>>> in > >>>>>>>>>> very rare edge cases: > >>>>>>>>>>>> > >>>>>>>>>>>> * UnaryOp now respects the target type. For doing that I had > to > >>>> also > >>>>>>>>>> change the Raw handling, finally fixing a bug that got > introduced > >> in > >>>>>>>> 2009 > >>>>>>>>>> ;) Before this fix all UnaryOps (SUM, MIN, MAX, CASE, etc) did > >>>> return > >>>>>>>> the > >>>>>>>>>> native type coming from the JDBC driver. That means that for a > >> TIME > >>>>>> WITH > >>>>>>>>>> TIME ZONE field we even did return vendor specific jdbc types > like > >>>>>>>>>> com.microsoft.jdbc.* or com.oracle.* types, etc. > >>>>>>>>>>>> > >>>>>>>>>>>> This mainly affects 2 areas: First, if there is a select sum, > >> max, > >>>>>>>> min, > >>>>>>>>>> case, etc which is used to return an Object[] and then cast up > to > >>>> the > >>>>>>>> type. > >>>>>>>>>> This might now fail, because we now return the correct type > >> defined > >>>> in > >>>>>>>> the > >>>>>>>>>> field. > >>>>>>>>>>>> E.g. if one did do a "select max(f.localDateTimeField) from > ..." > >>>>>> then > >>>>>>>>>> this used to return a driver specific type for many databases as > >>>>>>>> described > >>>>>>>>>> above. After the fix, we now return the type of the > >>>>>>>> 'localDateFimeField', > >>>>>>>>>> in this case java.time.LocalDateTime. > >>>>>>>>>>>> > >>>>>>>>>>>> Same happens for "select NEW" because right now we only look > >> for a > >>>>>>>>>> perfectly matching constructor and do no coercing. Should we > >>>> introduce > >>>>>>>>>> coercing probably? Means if a select new will result in a float > >>>> value > >>>>>>>> but > >>>>>>>>>> there is only a constructor for double, do we want to also > accept > >> it > >>>>>> in > >>>>>>>> the > >>>>>>>>>> future? > >>>>>>>>>>>> > >>>>>>>>>>>> * Along the way I also implemented BooleanRepresentation > >> handling > >>>>>> for > >>>>>>>>>> SQL literals via DBDictionary. > >>>>>>>>>>>> > >>>>>>>>>>>> > >>>>>>>>>>>> * respect TIMESTAMP precision in Oracle. Due to a bug we did > >>>>>> hardcoded > >>>>>>>>>> round at 3 digits precision. So we essentially only allowed > >> millis, > >>>>>>>> even on > >>>>>>>>>> a TIMESTAMP(6) field. The new code does respect the second > >> fractions > >>>>>> and > >>>>>>>>>> now defaults to 6. It should be compatible but it might behave > >> very > >>>>>>>> subtle > >>>>>>>>>> different. > >>>>>>>>>>>> > >>>>>>>>>>>> * fix the reserved column name handling by introducing > >>>>>>>>>> ColumnIdentifierRule (using the invalidColumnWordSet from the > >>>>>>>> DBDictionary > >>>>>>>>>> being used), separating it from the ColumnDefIdentifierRule. > >>>>>>>>>>>> > >>>>>>>>>>>> * fix SUM to always return Double as requested by the spec. > >>>>>> Previously > >>>>>>>>>> we did return whatever Numeric the JDBC driver did serve, > >> resulting > >>>> in > >>>>>>>> non > >>>>>>>>>> portable code. > >>>>>>>>>>>> > >>>>>>>>>>>> * PostgreSQL now supports setQueryTimeout. User might see this > >>>> come > >>>>>>>>>> alive and now return different when the situation occurs. > >>>>>>>>>>>> > >>>>>>>>>>>> > >>>>>>>>>>>> Does all that mean we should rather call the release 3.2.0 > >> rather > >>>>>> than > >>>>>>>>>> 3.1.3? > >>>>>>>>>>>> Or is the change so subtle that we still continue with 3.1.x? > >>>>>>>>>>> Hi Mark, > >>>>>>>>>>> first of all, thanks for your recent (hard) work to review and > >>>> close > >>>>>>>>>> long-standing issues. > >>>>>>>>>>> > >>>>>>>>>>> I don't have strong preference about versioning, but maybe > 3.2.0 > >>>>>> would > >>>>>>>>>> report more, to external, the idea of the amount of work done. > >>>>>>>>>>> > >>>>>>>>>>> Regards. > >>>>>>>>>>> > >>>>>>>>>>> -- > >>>>>>>>>>> Francesco Chicchiriccò > >>>>>>>>>>> > >>>>>>>>>>> Tirasa - Open Source Excellence > >>>>>>>>>>> http://www.tirasa.net/ <http://www.tirasa.net/> > >>>>>>>>>>> > >>>>>>>>>>> Member at The Apache Software Foundation > >>>>>>>>>>> Syncope, Cocoon, Olingo, CXF, OpenJPA, PonyMail > >>>>>>>>>>> http://home.apache.org/~ilgrosso/ < > >>>> http://home.apache.org/~ilgrosso/ > >
