Hi Vincent, The current behavior may be better to understand by bearing in mind that every execution of a BaseX command results in a new job that is orchestrated by the transaction manager. The TEST command behaves identically: Before it is started, it is ensured that no other job is running in parallel that would conflict with its execution. As we do not know in advance what tests will be performed (without analyzing all tests in more detail), tests will cause a global write lock. If a XQUnit test includes a job:execute call, it will be registered as new job and scheduled by the transaction management to be run after the test execution – which will never happen because everyone is waiting for good.
With the current solution, it is a logical consequence that XQUnit tests can also be run with the client/server architecture (but probably in a different way as you may have guessed): A client can initiate the test execution and the server will run the tests once there are no other jobs running. It is simply a consequence of defining test execution via a BaseX command, and I would assume this has rarely, if ever, been done in practice. One possible long-term change could be to define XQuery testing purely as a command-line based operation and ignore transactions completely (after all, there will never be any concurrent operations in a standalone command-line instance). A drawback of the drafted change would be that it would prevent us in future from introducing something like a unit:execute($url) function. But maybe this is over the top anyway. There are good reasons why we have no XQuery function to launch arbitrary BaseX commands: all this increases the danger of circular dependencies in the workflow, such as the one that we are currently discussing. How do you currently start the test execution? Christian ________________________________ Von: Lizzi, Vincent <[email protected]> Gesendet: Mittwoch, 4. Februar 2026 03:41 An: Christian Grün <[email protected]>; Christian Grün via BaseX-Talk <[email protected]> Betreff: Re: job:execute with unit:test stalls Hi Christian, Your explanation is helpful. The test scenarios that I have do need to read and write data. Currently it is possible to have a function annotated with %unit:before that performs updates to prepare databases for a %unit:test function. For example: module namespace t = "test"; declare %unit:before("t:test02") %updating function t:test01 () { db:create("test") }; declare %unit:before("t:test03") %updating function t:test02 () { db:put-value("test", 1 to 10, "numbers") }; declare %unit:test function t:test03 () { unit:assert(db:get-value("test", "numbers") = (1 to 10)) }; In a similar way, would it be possible to have a %unit:before function run code that uses job:execute, which may perform updates, and then have a %unit:test function verify the results, and avoid a deadlock? I'm not sure if this would be any easier than analyzing XQUnit functions to set a more fine granular lock as you described. I tried this and it resulted in a deadlock: module namespace t = "test"; declare %unit:before("t:test02") function t:test01 () { job:execute('db:create("test")') }; declare %unit:before("t:test03") function t:test02 () { job:execute('db:put-value("test", 1 to 10, "numbers")') }; declare %unit:test function t:test03 () { unit:assert(db:get-value("test", "numbers")) = (1 to 10) }; Perhaps a different approach could be to create tests using BaseX's client/server architecture: have XQUnit tests run in a BaseX client, and have code under test run in BaseX server facilitated by a set of RESTXQ functions. Many thanks, Vincent
