<< Back to previous view

[CLJ-1079] Don't squash explicit :line and :column metadata in the MetaReader Created: 29/Sep/12  Updated: 03/Sep/13

Status: Open
Project: Clojure
Component/s: None
Affects Version/s: Release 1.4, Release 1.5
Fix Version/s: None

Type: Defect Priority: Major
Reporter: Chas Emerick Assignee: Unassigned
Resolution: Unresolved Votes: 2
Labels: reader

Attachments: File CLJ-1079.diff    
Patch: Code and Test


I have been experimenting with using cljx to produce Clojure and ClojureScript source from a single file. This has gone well so far, with the exception that, due to the way the source transformation works, all of the linebreaks and other formatting is gone from the output. There is an option to include the original :line metadata in the output though, like so:

;;This file autogenerated from 
;;  src/cljx/com/foo/hosty.cljx
^{:line 1} (ns com.foo.hosty)
^{:line 3} (defn ^{:clj true} system-hash [x] ^{:line 5} (System/identityHashCode x))

(Hopefully, such hackery won't be necessary in the future with sjacket or something like it...)

Unfortunately, when read in using a LineNumberingPushbackReader, code like this has its :line metadata squashed by the line numbers coming from that. A REPL-friendly example would be:

=> (meta (read (clojure.lang.LineNumberingPushbackReader.
                 (java.io.StringReader. "^{:line 66} ()"))))
{:line 1}
=> (meta (read (java.io.PushbackReader.
                 (java.io.StringReader. "^{:line 66} ()"))))
{:line 66}

The latter seems more correct to me (and is equivalent to read-string).

Comment by Chas Emerick [ 29/Sep/12 7:07 PM ]

{{CLJ-1097.diff}} contains a fix for this issue, as well as a separate commit that eliminates a series of casts in order to improve readability in the area.

Comment by Andy Fingerhut [ 05/Oct/12 8:23 AM ]

Chas, your patch doesn't apply cleanly to latest Clojure master as of Oct 5 2012. I'm not sure, but I think some recent commits to Clojure on Oct 4 2012 caused that. I also take it as evidence of your awesomeness that you can write patches for tickets that haven't been filed yet

Comment by Chas Emerick [ 05/Oct/12 9:24 AM ]

"patches for tickets that haven't been filed yet?"

Anyway, tweaking up this patch is a small price to pay for having column meta. New {{CLJ-1097.diff}} patch attached, applies clean on master as of now. Otherwise same contents as in the original patch, except:

  • the same dynamic is also applied to :column metadata, now that it's available
  • the changes have been rebased into a single commit (including the elimination of the casts in MetaReader, which were becoming so numerous that the code was less readable than most
Comment by Nicola Mometto [ 05/Oct/12 9:39 AM ]

"patches for tickets that haven't been filed yet?"

He was referring to the fact that you uploaded "CLJ-1097.diff" while the ticket is #1079

Comment by Chas Emerick [ 05/Oct/12 9:42 AM ]

Oh, hah! Twice now, even! One more data point recommending my having slight dyslexia or somesuch. :-P

I've replaced the attached patch with one that is named properly to avoid any later confusion.

Comment by Chas Emerick [ 07/Oct/12 3:57 PM ]

Refreshed patch to apply cleanly to master after the recent off by one patch for :column metadata.

Comment by Stuart Halloway [ 19/Oct/12 3:12 PM ]

This feels backwards to me. If a special purpose tool wants to convey information via metadata, why does it use names that collide with those used by LispReader?

Comment by Chas Emerick [ 19/Oct/12 7:36 PM ]

The information being conveyed is the same :line and :column metadata conveyed by LispReader — in fact, that's where it comes from in the first place.

Kibit (and cljx) is essentially an out-of-band source transformation tool. Given an input like this:

(ns com.foo.hosty)

(defn ^:clj system-hash
 (System/identityHashCode x))

(defn ^:cljs system-hash
  (goog/getUid x))

…it produces two files, a .clj for Clojure, and a .cljs for ClojureScript. (The first code listing in the ticket description is the former.)

However, because there's no way to transform Clojure code/data without losing formatting, anything that depends on line/column numbers (stack traces, stepping debuggers) is significantly degraded. If LispReader were to defer to :line and :column metadata already available on the loaded forms (there when the two generated files are spit out with *print-meta* on), this would not be the case.

Comment by Andy Fingerhut [ 07/Feb/13 8:47 AM ]

clj-1079-patch-v2.txt dated Feb 7 2013 is identical to Chas's CLJ-1079.diff dated Oct 7 2012, except it applies cleanly to latest master. I believe the only difference is that some white space in the context lines is updated.

Comment by Andy Fingerhut [ 07/Feb/13 12:35 PM ]

Sorry for the noise. I've removed clj-1079-patch-v2.txt mentioned in the previous comment, because I learned that CLJ-1079.diff dated Oct 7 2012 applies cleanly to latest master and passes all tests if you use this command to apply it.

% git am --keep-cr -s --ignore-whitespace < CLJ-1079.diff

I will update the JIRA workflow page instructions for applying patches to mention this, too, because there are multiple patches that fail without --ignore-whitespace, but apply cleanly with that option. That will eliminate the need to update patches merely for whitespace changes.

Generated at Wed Oct 22 22:26:22 CDT 2014 using JIRA 4.4#649-r158309.