[ruby-core:85184] [Ruby trunk Bug#14413] `-n` and `-p` flags break when stdout is closed

Issue #14413 has been updated by nobu (Nobuyoshi Nakada).


A patch to exit with SIGPIPE when `EPIPE` if `-n` or `-p` option is given.

```diff
diff --git i/error.c w/error.c
index 7cbbf101e0..2f454d01da 100644
--- i/error.c
+++ w/error.c
@@ -53,6 +53,7 @@ int rb_str_end_with_asciichar(VALUE str, int c);
 VALUE rb_eEAGAIN;
 VALUE rb_eEWOULDBLOCK;
 VALUE rb_eEINPROGRESS;
+VALUE rb_eEPIPE;
 static VALUE rb_mWarning;
 static VALUE rb_cWarningBuffer;
 
@@ -1816,6 +1817,9 @@ set_syserr(int n, const char *name)
 	  case EINPROGRESS:
 	    rb_eEINPROGRESS = error;
 	    break;
+	  case EPIPE:
+	    rb_eEPIPE = error;
+	    break;
 	}
 
 	rb_define_const(error, "Errno", INT2NUM(n));
diff --git i/parse.y w/parse.y
index c8acac5d2c..57dbc7f7db 100644
--- i/parse.y
+++ w/parse.y
@@ -10571,11 +10571,13 @@ rb_parser_warn_location(VALUE vparser, int warn)
     p->warn_location = warn;
 }
 
+extern VALUE rb_eEPIPE;
 static NODE *
 parser_append_options(struct parser_params *p, NODE *node)
 {
     static const YYLTYPE default_location = {{1, 0}, {1, 0}};
     const YYLTYPE *const LOC = &default_location;
+    int ignore_epipe = p->do_print || p->do_loop;
 
     if (p->do_print) {
 	NODE *print = NEW_FCALL(rb_intern("print"),
@@ -10584,6 +10586,19 @@ parser_append_options(struct parser_params *p, NODE *node)
 	node = block_append(p, node, print);
     }
 
+    if (ignore_epipe) {
+	/* rescue Errno::EPIPE then raise SignalException.new(SIGPIPE) */
+	VALUE signo = INT2FIX(SIGPIPE);
+	VALUE exc = rb_class_new_instance(1, &signo, rb_eSignal);
+	NODE *res = NEW_RESBODY(NEW_LIST(NEW_LIT(rb_eEPIPE, LOC), LOC),
+				NEW_FCALL(rb_intern("raise"),
+					  NEW_LIST(NEW_LIT(exc, LOC), LOC),
+					  LOC),
+				0,
+				LOC);
+	node = NEW_RESCUE(node, res, 0, LOC);
+    }
+
     if (p->do_loop) {
 	if (p->do_split) {
 	    NODE *split = NEW_GASGN(rb_intern("$F"),
```

----------------------------------------
Bug #14413: `-n` and `-p` flags break when stdout is closed
https://bugs.ruby-lang.org/issues/14413#change-69929

* Author: josh.cheek (Josh Cheek)
* Status: Feedback
* Priority: Normal
* Assignee: 
* Target version: 
* ruby -v: ruby 2.5.0preview1 (2017-10-10 trunk 60153) [x86_64-darwin16]
* Backport: 2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: UNKNOWN
----------------------------------------
Ruby generally works well within a pipeline. The `-n` and `-p` flags are incredibly useful. However, it is common practice to use programs like `head` and `sed`, which will close the pipe after completing their job. This is convenient, because often it limits an expensive amount of output to a workable subset. I can figure out the pipeline for a subset of input, and then remove the limiting function.

However, Ruby explodes with `-e:1:in `write': Broken pipe @ io_write - <STDOUT> (Errno::EPIPE)`, when it writes the current line to stdout. When in a line oriented mode, and stdout closes, I think it should exit successfully (so it doesn't break bash scripts with `pipe fail`set) and silently (no error message), hence marking this a bug report rather than a feature request.

I've attached a screenshot to show that this is how every other program works, that I tried.

~~~
git clone https://github.com/jquery/esprima
cd esprima/test/fixtures/
ls                                                                             | head -1 # ls
find . -type f -name '*json'                                                   | head -1 # find
find . -type f -name '*json' | head -1 | xargs cat | jq .                      | head -1 # jq
find . -type f -name '*json' | grep -v JSX | grep -v tokenize                  | head -1 # grep
find . -type f -name '*json' | sed -E '/JSX|tokenize/d'                        | head -1 # sed
find . -type f -name '*json' | awk '/JSX|tokenize/ { next }; { print }'        | head -1 # awk
find . -type f -name '*json' | perl -e 'while(<>) { /JSX|tokenize/ || print }' | head -1 # perl
find . -type f -name '*json' | ruby -ne 'print unless /JSX|tokenize/'          | head -1 # ruby :(
~~~

---Files--------------------------------
Screen Shot 2018-01-27 at 7.27.49 PM.png (458 KB)


-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>