Hi Mary, Thank you for this clear explanation. This is excellent news (news to me at least), I always assumed try/catch were trickier than that WRT errors.
Just out of curiosity, how is that possible efficiently? I can only imaging using several MVCC copies, that we create when entering a try block, and on exiting the block we use either the new copy, or the old one in case of an error... Regards, -- Florent Georges H2O Consulting http://h2o.consulting/ On 11 December 2017 at 16:27, Mary Holstege wrote: > > Updates within the body of the try will be abandoned; updates before or > after will be kept. That is, as long as they are catchable exceptions. If > you get an exception in the commit of the whole transaction, all > transactions will be lost. > > That's as far as updates go. Some side-effects are not transactionally > protected, such as writing to a log file. > > //Mary > > > On Mon, 11 Dec 2017 05:17:00 -0800, Florent Georges <[email protected]> > wrote: > > > Hi, > > > > I would like to have a confirmation about the handling of side effects > > (that is, everything related to MVCC) in the context of try/catch. I > > always thought that "whatever was executed in a try clause" before an > > error > > is raised was going to be kept if the error is caught. > > > > I know there is an incertainty WRT the order of execution, but still, I > > thought some DB modifications called in the try clause could be kept in > > case the error is caught. > > > > Here is an example that tries to replicate the problem, using a recursive > > function to beat the optimiser (full code at the end of this email): > > > > local:side-effect('before'), > > try { > > local:recurse(('one', 'two')) > > } > > catch * { > > 'Caught error in main module' > > }, > > local:side-effect('after') > > > > In the above example, the result is that both /before.xml and /after.xml > > are created, but not /one.xml (the creation is called during the first > > iteration of the recursive function though). > > > > All the tests I made have the same result: all DB modifications in the > > try > > clause are discarded in case an error is raised (regardless it is caught > > or > > not). > > > > I could not find a definitive answer in the doc. Is this guaranteed? > > > > Full example: > > > > xquery version "3.0"; > > > > declare namespace xdmp = "http://marklogic.com/xdmp";; > > > > declare function local:side-effect($name as xs:string) > > { > > xdmp:document-insert( > > '/' || $name || '.xml', > > <hello>{ $name }</hello>) > > }; > > > > declare function local:recurse($names as xs:string) > > { > > if ( fn:count($names) eq 1 ) then ( > > fn:error((), 'Climb the stack...') > > ) > > else ( > > local:side-effect(fn:head($names)), > > local:recurse(fn:tail($names)) > > ) > > }; > > > > local:side-effect('before'), > > try { > > local:recurse(('one', 'two')) > > } > > catch * { > > 'Caught error in main module' > > }, > > local:side-effect('after') > > > > Regards, > > > > > -- > Using Opera's revolutionary email client: http://www.opera.com/mail/ > _______________________________________________ > General mailing list > [email protected] > Manage your subscription at: > http://developer.marklogic.com/mailman/listinfo/general >
_______________________________________________ General mailing list [email protected] Manage your subscription at: http://developer.marklogic.com/mailman/listinfo/general
