I have a feature request which I was going to ask for in meta.math.SE as well, but figured this place would benefit from it too, while the SO developers will be less inclined to implement it (I'd think that it is lower in the priority order)

Much like the "cite" export function, it could be nice if there was an "import" function in which we can specify a DOI and the object will be instead given a nice citation form?

Since DOIs are pretty static, this can be an external (or not) script to the SE program which will hardcode the resolved data into the post instead of leaving a DOI tag to be resolved every time the post is loaded.

How does that sound?

Okay, I will clarify.

First the way I see it you can either hack into the SE software and add, perhaps, another markup symbol; or you could have an external code that will recognize DOI objects (or will be fed DOI code directly) and change them into something else.

Into what? Into a predefined citation + link. For example,

Howard, P. The axiom of choice for countable collections of countable sets does not imply the countable union theorem. Notre Dame J. Formal Logic Volume 33, Number 2 (1992), 236-243.

Perhaps the link to the journal can be taken off, but the point is that we replace something like doi(...) with a bibliographic+link entry of text.

The part about static DOI code is that unlike links to somewhere (including, for example, MO) the DOI codes are not expected to change anytime soon. Therefore if we prefer to have it parse the DOI code every time (instead once) then the entry is not bound to change. A certain code will remain there for the foreseeable future.

At present, and very likely in the future, our ability to implement features like this will be limited to client-side javascript.

Thus there are two options:

1. intercept the submission of post text, performing the appropriate lookup and replacement.
2. rewrite the displayed post text each time.

Both are possible, although the second one relies on all future viewers of the content running the appropriate javascript. (Of course, using LaTeX via MathJax suffers from the same problem.) Both can easily be implemented as a userscript. If someone were to offer a nice implementation via option 1), I might be in favour of turning it on by default for all MO users. I'm not personally interested in doing the implementation; it strikes me as something that will receive relatively little use.

@José,

nice MediaWiki extension! I like the idea of nicely handling arxiv links better than nicely handling arbitrary DOIs: in particular, as far as I know there's no way to query a DOI for any of the metadata (e.g. the title).

@Scott: I did not know that there is no uniform way to fetch DOI metadata. I believe they might be working on such project since the doi.org website says that there is doi API in a private beta stage.

Like Jose I too think that a uniform standard for citation on the site would be a very good thing. As far as I know arXiv has DOI strings, so a DOI interface would be useful for arXiv papers too.

Update: http://www.crossref.org/guestquery/ has an XML output of metadata. It seems to be possible after all.

In a short correspondence with CrossRef it turns that one needs either become a member (I'd assume of CrossRef, I don't know how much it would cost to have MO with access to the site); or one can use OpenURL for that. I am not sure how exactly, though.

I'm with @quid here. MathSciNet is great and all, but I think it's a bit rude to give people links to paywall'd links if you're not sure they'll have access.

@Willie: agreed that ZB larger size is (mainly, or perhaps only) due to its longer existance, but still I have not heard which relevant items are (supposedly?) in MR and not ZB. Indeed, in my experience for MO questions references to old papers are not irrelevant (not too few questions deal with things like when was this or that introduced, who proved what exactly, and so on; going sometimes well back to the 1930s and beyond); and rather more relevant than things sufficiently non-main-stream to be not in both. What is true is that MR deals relatively gracefully with getitem-links in the absence of a subscription. However, I am worried that if people develop the habit of linking to MR many will start to implicitly assume that then everybody has the review/abstract handy and assume this in their answers/questions, which as it is not the case could become a problem. [If for some reason it turns out to be necessary to link to MR it is also now not at all hard, and indeed not much harder than with support. I mean the MR or Zbl number is about the last thing I know by heart about a paper; so if I know the MR number I typically have everything else handy too, and in all likelihood even the relevant page open.]

@WillieWong and @Dmitri, thanks, I'd forgotten about the relay mechanism. I agree this is fine.

I have no concrete suggestion, but one point is important to me. And, this perhaps starts a step earlier. Namely, what is the actual goal, or what should 'handle nicely' mean.

Asaf's original suggestion was such that using the mechanism would tend to make the MO-post more self-contained; the post would contain full bib-details.

I would dislike anything that invites making posts less self-contained. For example I would dislike a simple solution like turning !MR123456 into a MR123456 plus linking to the relevant MathSciNet review via getitem; same for arXiv. (This is othogonal to the question whether the full content is visible for everybody or not.)

I strongly prefer if I can get at least a rough idea (better a precise one, though I know I do not always comply to this when writing) what is linked to whithout clicking through. By contrast, if I have full bib-details a stable link to the official version, the MathSciNet page, the arXiv page,... to me seem like a very minor convenience, basically irrelevant, since these are trivial to find for everybody anyway. In particular, if there are no plans with the MO content I am unanware of and I am not missing something, consistent references do not seem important to me; they would be nice and one could imagine making use of this in one form or another, but not important (the questions are so loosely related that whether Question 1234 and Question 11234 link to the same thing in a different or the same way seems not so important to me). Yet, that posts are self-contained/convenient to read is very important to me.

I am not sure this will go into this direction, but preemptively: please let us not make a standard, or invite doing, something that is (at least close to the time of writing, that is when most people will read) worse than posting a 'naked' link from which I sometimes can infer much more than from MR123456.

In my opinion, a 'simple solution' could be worse than not doing anything. And, then I definitely think the latter would be better.

p.s. I am aware that there are plenty of links of the form 'here', but the question is what is arround, and I believe that there is a psychological difference when writing. When writing: 'see here' it is clear that this is not self-contained and one might thus not do this without giving some context. While 'see MR123456' might give the illusion of being self-contained.

@Asaf: Yes, I know that you did not have this idea, so I gave your suggestion as an in my opinion positive example. I do not know the ideas of 'anyone'. Dimiti Pavlov suggested "we consider adding a standardized way to refer to MathSciNet" I do not know what the precise suggestion is. But from the further discussion it seems to be about linking to not querying and extracting information from MathSciNet to include it in the post. And if you do not query what else but the number do you have.

You say (emphasize mine): "I believe that if writing [MR123456] will automatically be resolved into a full citation with a link it will increase the number of users that add these citations. I like that in my posts and I like that in other's posts."

Yes that would be convenient. But I was not sure everybody was talking about this, as opposed to only a link. And, my point is that support for only a link (which I imagine would be quite straightforward to do) would be worse than nothing, and not a small step in the right direction.

I am glad to hear that you are convinced everybody agrees on this.

@Dmitri: we could even add the actual PDF file to the links for those of us outside the EMS paywall :)

@Dmitri: I know. I was just curious about the paper you mentioned, couldn't get past the paywall and found it in Teichner's website. I didn't mean to imply you could get it algorithmically -- hence the :)