Validation happens in the API, so it applies equally to the API, the webapp, and web services.
Validation happens whenever you save the parent object (e.g. VisitService.saveVisit does validate(visit); dao.save(visit).) And the only way to interact with these child objects (e.g. Visit Attributes) is through the parent (e.g. we don't do provide a saveVisitAttribute method). So doing validation at the time of saving the parent is "safe". The design pattern we try to follow is that the domain objects are "dumb", and the intelligence in in the API (which includes the validators). Also, the addAttribute method can't do full validation because the domain object (e.g. visit) may be in an invalid state between saves. Burke, Roger, my idea is that "addAttribute" is a plain method that does exactly what you'd expect. Whereas setAttribute is a convenience method documented to "Sets an attribute, voiding any existing attributes on the holder of the same attribute type. Will fail if the relevant attribute type requires more than one value." Roger, I don't understand the bit about double iteration, and Doctor of the Year for different years. -Darius On Sun, Sep 25, 2011 at 7:50 PM, Burke Mamlin <[email protected]>wrote: > It might be cleaner to handle repeats either (1) within the datatype or (2) > through a separate interface (e.g., RepeatingCustomDataType) that would have > add/remove methods for single values and use List<T> for get/set. > > Trying to manage get/set vs. add/remove within a single interface, as Roger > suggests, could make things more complicated than necessary. > > -Burke > > > On Sun, Sep 25, 2011 at 6:03 PM, Friedman, Roger (CDC/CGH/DGHA) (CTR) < > [email protected]> wrote: > >> I have forked this part of the thread Empty String Fields v. Null >> Fields.**** >> >> ** ** >> >> You are assuming that all class updates come through form controllers. >> But we have updates coming from REST posts and bulk load processes as well. >> So I think the validation has to be at the API level (perhaps the form >> controller as well).**** >> >> ** ** >> >> I don’t know that you’ve got the double iteration (over attribute values >> within attribute types) down quite right. AFAIK, attribute values for any >> one attribute for any one object constitute a set; we can’t have a “Doctor >> of the Year” award, we have to have “Doctor of the Year 2001”, “Doctor of >> the Year 2002”, etc. to populate the provider object’s awards attribute if >> Burke is to win it every year. **** >> >> ** ** >> >> Surely we still have methods like visit_attribute.setAttribute(value) or >> visit.attributes(x).setAttribute(value); we’d need to do the void if null >> thing; we’d also have to do the void thing if the value is being set to the >> value of another existing attribute (or we could throw an exception and let >> the caller void the attribute value with an appropriate reason).**** >> >> ** ** >> >> I don’t see how visit.setAttribute(attrType, value) works. Is that method >> included for compatibility’s sake and throws an exception for attribute >> types whose multiplicity is not (0,1) (or whose attribute type count is not >> 0 or 1)? Or are you saying that visit.addAttribute adds another attribute >> of type attrType while visit.setAttribute voids all existing attributes of >> type attrType for that visit and adds a new attribute with the given value? >> Seems like a lot of extra voiding. What’s its inverse operation – >> visit.getAttribute(attrType)? what does it return, maybe an iterator?** >> ** >> >> ** ** >> >> Duplicate detection and voiding when null seem both to be doable when the >> add/set call is made; the count will be available due to duplicate >> detection, so we could check multiplicity then. I understand your wanting >> to do validation at the object rather than the object-property level, but >> the error detection could come a long way from error introduction. Maybe >> the object needs to store a working copy of the attribute ids and values and >> void reasons in addition to the attribute collection, let the attribute >> changes/validations affect the working copy, then have the save object >> method actually update the hibernate objects in the attribute collection >> from the working copy.**** >> >> ** ** >> >> *From:* [email protected] [mailto:[email protected]] *On Behalf Of *Darius >> Jazayeri >> *Sent:* Sunday, September 25, 2011 2:19 PM >> *To:* [email protected] >> *Subject:* Re: [OPENMRS-DEV] Empty string fields vs null fields**** >> >> ** ** >> >> Multiplicity constraints on attributes are enforced by the validator of >> the class that owns those attributes. I.e. if you have a VisitAttributeType >> with minOccurs=1 and maxOccurs=3, that will be checked when you validate the >> Visit it belongs to.**** >> >> ** ** >> >> I'd hope things would work like this:**** >> >> (1) visit.setAttribute(attrType, null) --> semantically equivalent to >> "void any existing attributes of the given type"**** >> >> (2) visit.addAttribute(attrType, null) --> this should fail when you try >> to save the visit, because a null-valued attribute is invalid. (The db has a >> not-null constraint on the value column.) To get a pretty error in this >> case, we'd probably need to handle this in the controller for the edit visit >> page. But at least the AOP would prevent us from storing empty values for >> attributes (unintentionally sidestepping the not-null constraint.)**** >> >> ** ** >> >> -Darius**** >> >> ** ** >> >> On Sun, Sep 25, 2011 at 8:09 AM, Friedman, Roger (CDC/CGH/DGHA) (CTR) < >> [email protected]> wrote:**** >> >> I actually agree that something should be put in the field, but I have >> seen so many variants (blank, ?, ??, ???, Unknown, Uknown, Unown, Not Known, >> Not Available, Missing, Left Card At Home) that I have come to believe that >> the easiest thing to say to users is “if you don’t know it, leave it >> blank”. The form controller normally will know whether an unknown value is >> required and if so what its representation should be; however, that doesn’t >> apply to our 3 formentry modules. In the case of OpenMRS objects, we could >> use ID=-1 to indicate unknown, perhaps providing a default entry in the >> underlying table (to permit inner rather than left joins).**** >> >> **** >> >> But you still haven’t said where the multiplicity constraints on >> attributes should be enforced and whether that’s compatible with using AOP >> to convert empty strings to nulls and what the effect of updating an >> attribute value to null should be. **** >> >> **** >> >> *From:* [email protected] [mailto:[email protected]] *On Behalf Of *Darius >> Jazayeri >> *Sent:* Friday, September 23, 2011 2:44 PM**** >> >> >> *To:* [email protected] >> *Subject:* Re: [OPENMRS-DEV] Empty string fields vs null fields**** >> >> **** >> >> I agree with Burke here, I think that allowing people to store "" in a >> required field, and using that to represent "yes, but unknown" isn't right. >> You should store an explicit and recognizable unknown value, e.g. "Unknown". >> So for the guardian case, no guardian means do not store a row in the >> person_attribute table for the non-existent guardian. And an >> existent-but-unknown guardian would be stored as an attribute whose value is >> a pointer to an "Unknown Person" person.**** >> >> **** >> >> Also, at a web UI level, the standard widgets don't distinguish between >> these things (i.e. if you have an attribute type whose datatype is "person", >> the out-of-the-box widget is just going to show a person chooser, not have >> an extra "click for unknown")...**** >> >> **** >> >> -Darius**** >> >> >> On Friday, September 23, 2011, Burke Mamlin wrote:**** >> >> Roger,**** >> >> **** >> >> As you suggest, treating all empty strings as null will enforce that >> required fields cannot be blank, since we won't be able to distinguish >> between answered with an empty string from not answered at all. That said, >> when is it "okay" to respond to a required question with nothing? Wouldn't >> your example be better handled by recording the "Unknown" (in the case that >> the orphan has a legal guardian who is currently unknown) instead of >> treating unknown the same as if the field were skipped? And/or record "N/A" >> when the field doesn't apply (in OpenMRS, we would probably just omit >> storing an observation altogether if it didn't apply).**** >> >> **** >> >> -Burke**** >> >> On Fri, Sep 23, 2011 at 8:59 AM, Friedman, Roger (CDC/CGH/DGHA) (CTR) < >> [email protected]> wrote:**** >> >> Darius –**** >> >> I believe I gave several examples during the earlier round of this >> discussion. Here’s one – an orphan has a legal guardian, but the person >> bringing the orphan to the clinic doesn’t know who, so the field is left >> blank; a non-orphan does not have a legal guardian. So when it comes time >> to report, you do something patient left join person_attribute where >> attribute_type=”Legal Guardian” and format the nulls as “N/A” and print the >> empty strings.**** >> >> I don’t think the automagic works with the new attribute paradigm. >> Take a required attribute like National ID. The patient does have her card >> with her when she comes to the clinic. Now we can’t save the patient info. >> So I withdraw my tentative conclusion about properties. Is the >> implementation of the minimum constraint still undecided? Let’s make sure >> that’s well-understood before moving on to automagic.**** >> >> **** >> >> *From:* [email protected] [mailto:[email protected]] *On Behalf Of *Darius >> Jazayeri >> *Sent:* Thursday, September 22, 2011 4:36 PM**** >> >> >> *To:* [email protected] >> *Subject:* Re: [OPENMRS-DEV] Empty string fields vs null fields**** >> >> **** >> >> https://tickets.openmrs.org/browse/TRUNK-2680**** >> >> On Thu, Sep 22, 2011 at 1:24 PM, Darius Jazayeri <[email protected]> >> wrote:**** >> >> There is already AOP happening on every XyzService.saveXyz call. So we're >> actually already paying the significant overhead. I think that >> reflection+iterating over properties only on saveXyz methods is vanishingly >> small additional overhead to the fact that every service call is AOPd.*** >> * >> >> **** >> >> -Darius**** >> >> **** >> >> On Thu, Sep 22, 2011 at 12:51 PM, Burke Mamlin <[email protected]> >> wrote:**** >> >> On Thu, Sep 22, 2011 at 3:03 PM, Darius Jazayeri <[email protected]> >> wrote:**** >> >> So, does that mean you're fine with automagically changing empty String >> properties of OpenmrsObjects to null before saving to the database?**** >> >> **** >> >> Yes. I love magic (I'm no muggle!)… as long as it's not going to add >> several milliseconds or more to every API call. If doing it automagically >> involves AOP or some other hook that is going to add (or contribute to) a >> measurable performance hit, I would rather find a way to manage it >> explicitly (via hibernate configuration, a custom hibernate type, and/or >> code reviews).**** >> >> **** >> >> -Burke >> > > ------------------------------ > Click here to > unsubscribe<[email protected]?body=SIGNOFF%20openmrs-devel-l>from > OpenMRS Developers' mailing list > _________________________________________ To unsubscribe from OpenMRS Developers' mailing list, send an e-mail to [email protected] with "SIGNOFF openmrs-devel-l" in the body (not the subject) of your e-mail. [mailto:[email protected]?body=SIGNOFF%20openmrs-devel-l]
