http://dev.clojure.org/jira This file is an XML representation of an issue en-us 4.4 649 25-07-2011 http://dev.clojure.org/jira/browse/CLJ-733 Clojure <p>Problem: need to convey diagnostic information in an exception, and Java exceptions are poor for this because they don't have any place to put data. Having many apps write their own exceptions is unnecessary code bloat, and sometimes has compilation implications.</p> <p>Choices made:</p> <ul> <li>Naming it <tt>ExceptionInfo</tt>. Needs to tell the user that it carries data, without suggesting that there was a problem with data. <ul> <li>think we have to include Exception in the name, esp. for interop consumers.</li> </ul> </li> <li>Class of data member. IPersistentMap</li> <li>Class of map input. IPersistentMap</li> <li>Do not want to build anything to support conditional logic. (Use bindings for this).</li> <li>Base class. RuntimeException is consistent with a dynamic language (don't want to impose checked exceptions).</li> <li>print with data just after msg on first line</li> <li>equality by identity (pretty much dictated by having exception as base class)</li> <li>ex-info factory function</li> <li>ex-data fn to get the data out</li> <li>data and msg arguments required</li> <li>data argument comes second <ul> <li>matches Java expectation of msg first, cause last</li> <li>matches print order</li> </ul> </li> </ul> CLJ-733

Data Conveying Exception

Enhancement Minor Closed Completed Stuart Halloway Stuart Halloway Fri, 28 Jan 2011 13:59:41 -0600 Fri, 9 Dec 2011 10:01:00 -0600 Fri, 9 Dec 2011 10:01:00 -0600 Approved Backlog 0 0 <p>Looks good; Stu's patch was missing a forward declaration to print-defrecord. 0733-data-conveying-exception-2.patch adds one, and all tests pass.</p> <p>Looks like is-a to me. Why is exception a map?</p> <p>Also, bad practice to merge non-namespace-qualified keys with something you don't originate - should have ns. Actually, why merge at all? Why not let higher-level exception reporting handle cause and message?</p> <p>Is there a convention for nesting these, merging data?</p> <p>type checks for :type in metadata, might be nice if that worked on DCE's, maybe they should support metadata</p> <p>Latest patch per discussion on IRC <a href="http://clojure-log.n01se.net/#11:12b">http://clojure-log.n01se.net/#11:12b</a> . Note that I had to touch string representation in three places (Java, main, and repl). I think main is only used for repl anyway, so a future cleanup could have main and repl share printing code.</p> <p>Ctor arg order is a question. </p> <p>Also, do we want to encourage making an exception w/o a message?</p> <p>I think it should have been ex-data, in my last line on IRC, not ex-info</p> <p>Will we have a fn to make these rather than encode the type via ctor call everywhere? Maybe that should be called ex-info. Same fn could be useful in cljs</p> <p>Also the description no longer describes the approach</p> <p>Been back and forth on argument order, but I think msg/data/cause is best.</p> <p>Only issue I think is whether you are going to enforce a map (and not nil) be passed. Currently you don't, making your doc string for ex-data wrong - it returns .getData of ExceptionInfo objects, which may be nil. If we enforce non-nil map on the way in, then ex-data works as documented and people can catch the exception and branch on ex-data. That seems most useful.</p> <p>Also please mark as alpha, subject to change.</p> <p>Attached <tt>0733-with-map-check.patch</tt>. Same as <tt>0733-with-ex-data-and-ex-info.patch</tt> but added Alpha docstring and map check to <tt>ex-data</tt> per Rich's latest comment.</p> <p>Thanks. The nil check should probably go in the ExceptionInfo ctor, just to have the same enforcement should they be created from Java.</p> <p>I think this line in <tt>clojure.repl/pst</tt>,</p> <div class="code panel" style="border-width: 1px;"><div class="codeContent panelContent"> <pre class="code-java">(when-let [info (ex-data e)] (str <span class="code-quote">" "</span> info))</pre> </div></div> <p>should be using <tt>pr-str</tt> instead of <tt>str</tt>, although <tt>str</tt> on a map already quotes string keys/values, so maybe it's good enough.</p> <p>Setting to 'Incomplete' pending response to 2 previous comments.</p> <p>The dec 2 edition does the arg check in the constructor and uses pr-str, addressing the last two comments raised.</p> <p>The dec 2 edition does the arg check in the constructor and uses pr-str, addressing the last two comments raised.</p> <p>Screened &amp; moved to Approved Backlog.</p> <p>ex-info and ex-data seem rather terse names to me (exception-info and exception-data would be clearer, imho). Clojure seems to have a mixture of abbreviated names and full names, but it was my vague impression that most of the abbreviations stemmed from traditional lisp or java usage.</p> Approval Ok Global Rank Patch Code and Test