[PATCH] Handle line continuation in unquoted heredocs

[PATCH] Handle line continuation in unquoted heredocs

[Denys Vlasenko]

Testcases:

cat <<nice
heredoc1
ni\
ce
nice # this isn't heredoc end, the end is above

cat <<-nice
	heredoc2
	\
	ni\
ce
nice # this isn't heredoc end, the end is above

cat <<"nice"
heredoc3
ni\
ce
nice

cat <<-"nice"
	heredoc4
	\
	ni\
ce
	nice

Correct output:

heredoc1
0
heredoc2
0
heredoc3
ni\
ce
heredoc4
\
ni\
ce

Before patch, cases 1 and 2 are mishandled.

A more simplistic apparent solution for cases 1 and 2,
s/pgetc/pgetc_eatbnl/, breaks cases 3 and 4

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
---
 src/parser.c | 33 ++++++++++++++++++++-------------
 1 file changed, 20 insertions(+), 13 deletions(-)

diff --git a/src/parser.c b/src/parser.c
index aecc18f..edef0ba 100644
--- a/src/parser.c
+++ b/src/parser.c
@@ -74,7 +74,7 @@
 
 
 /* Used by expandstr to get here-doc like behaviour. */
-#define FAKEEOFMARK (char *)1
+#define FAKEEOFMARK (struct heredoc *)1
 
 
 
@@ -123,7 +123,7 @@ STATIC void parseheredoc(void);
 STATIC int readtoken(void);
 STATIC int xxreadtoken(void);
 STATIC int pgetc_eatbnl();
-STATIC int readtoken1(int, char const *, char *, int);
+STATIC int readtoken1(int, char const *, struct heredoc *);
 STATIC void synexpect(int) __attribute__((__noreturn__));
 STATIC void synerror(const char *) __attribute__((__noreturn__));
 STATIC void setprompt(int);
@@ -143,7 +143,7 @@ int issimplecmd(union node *n, const char *name)
 	       equal(n->ncmd.args->narg.text, name);
 }
 
-static inline int realeofmark(const char *eofmark)
+static inline int realeofmark(struct heredoc *eofmark)
 {
 	return eofmark && eofmark != FAKEEOFMARK;
 }
@@ -685,9 +685,9 @@ parseheredoc(void)
 			setprompt(2);
 		}
 		if (here->here->type == NHERE)
-			readtoken1(pgetc(), SQSYNTAX, here->eofmark, here->striptabs);
+			readtoken1(pgetc(), SQSYNTAX, here);
 		else
-			readtoken1(pgetc_eatbnl(), DQSYNTAX, here->eofmark, here->striptabs);
+			readtoken1(pgetc_eatbnl(), DQSYNTAX, here);
 		n = (union node *)stalloc(sizeof (struct narg));
 		n->narg.type = NARG;
 		n->narg.next = NULL;
@@ -840,7 +840,7 @@ xxreadtoken(void)
 		case ')':
 			RETURN(TRP);
 		}
-		tok = readtoken1(c, BASESYNTAX, (char *)NULL, 0);
+		tok = readtoken1(c, BASESYNTAX, (struct heredoc *)NULL);
 		if (tok != TBLANK)
 			return tok;
 	}
@@ -994,7 +994,7 @@ static char *dollarsq_escape(char *out)
 #define	PARSEARITH()	{goto parsearith; parsearith_return:;}
 
 STATIC int
-readtoken1(int firstc, char const *syntax, char *eofmark, int striptabs)
+readtoken1(int firstc, char const *syntax, struct heredoc *eofmark)
 {
 	struct synstack synbase = { .syntax = syntax };
 	int chkeofmark = checkkwd & CHKEOFMARK;
@@ -1233,17 +1233,24 @@ checkend: {
 		int markloc;
 		char *p;
 
-		if (striptabs) {
-			while (c == '\t')
-				c = pgetc();
+		if (eofmark->striptabs) {
+			while (c == '\t') {
+				if (eofmark->here->type == NHERE)
+					c = pgetc();
+				else /* NXHERE */
+					c = pgetc_eatbnl();
+			}
 		}
 
 		markloc = out - (char *)stackblock();
-		for (p = eofmark; STPUTC(c, out), *p; p++) {
+		for (p = eofmark->eofmark; STPUTC(c, out), *p; p++) {
 			if (c != *p)
 				goto more_heredoc;
 
-			c = pgetc();
+			if (eofmark->here->type == NHERE)
+				c = pgetc();
+			else /* NXHERE */
+				c = pgetc_eatbnl();
 		}
 
 		if (c == '\n' || c == PEOF) {
@@ -1748,7 +1755,7 @@ expandstr(const char *ps)
 		goto out;
 	handler = &jmploc;
 
-	readtoken1(pgetc_eatbnl(), DQSYNTAX, FAKEEOFMARK, 0);
+	readtoken1(pgetc_eatbnl(), DQSYNTAX, FAKEEOFMARK);
 
 	n.narg.type = NARG;
 	n.narg.next = NULL;
-- 
2.41.0

Re: [PATCH] Handle line continuation in unquoted heredocs

[Harald van Dijk]

On 07/08/2025 21:10, Denys Vlasenko wrote:
> Testcases:
> 
> cat <<nice
> heredoc1
> ni\
> ce
> nice # this isn't heredoc end, the end is above

Interesting. I am having trouble telling from the POSIX wording whether 
this is meant to be valid. Is this the case that "It is unspecified 
whether the line containing the trailing delimiter is itself subject to 
this line continuation." is meant to apply to, or not?

> A more simplistic apparent solution for cases 1 and 2,
> s/pgetc/pgetc_eatbnl/, breaks cases 3 and 4

Would s/pgetc/pgetc_top(synstack)/ work instead?

Cheers,
Harald van Dijk

Re: [PATCH] Handle line continuation in unquoted heredocs

[Herbert Xu]

Denys Vlasenko <dvlasenk@redhat.com> wrote:
> Testcases:
> 
> cat <<nice
> heredoc1
> ni\
> ce
> nice # this isn't heredoc end, the end is above

I think this is debatable.  The current behaviour differs from
bash, but matches ksh.

Since POSIX apparently left this as unspecified, I'm not changing
dash.

Thanks,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt