From e75dbb6af8fbea53c62efb7176ed2d25a47557c9 Mon Sep 17 00:00:00 2001
From: Russ Cox <rsc@swtch.com>
Date: Wed, 15 Jan 2020 10:48:20 -0500
Subject: factotum: update for new nbrecvul return value

Unclear whether the old semantics were the right ones,
but at least this preserves what they've been for the past
however many years.
---
 src/cmd/auth/factotum/confirm.c | 32 ++++++++++++++++++++++++++++++--
 1 file changed, 30 insertions(+), 2 deletions(-)

(limited to 'src')

diff --git a/src/cmd/auth/factotum/confirm.c b/src/cmd/auth/factotum/confirm.c
index 105a46db..8451e8ae 100644
--- a/src/cmd/auth/factotum/confirm.c
+++ b/src/cmd/auth/factotum/confirm.c
@@ -128,17 +128,33 @@ needkeywrite(char *s)
 int
 needkey(Conv *c, Attr *a)
 {
+	ulong u;
+
 	if(c == nil || *needkeyinuse == 0)
 		return -1;
 
 	lbappend(&needkeybuf, "needkey tag=%lud %A", c->tag, a);
 	flog("needkey %A", a);
-	return nbrecvul(c->keywait);
+
+	// Note: This code used to "return nbrecvul(c->keywait)."
+	// In Jan 2020 we changed nbrecvul to match Plan 9 and
+	// the man page and return 0 on "no data available" instead
+	// of -1. This new code with an explicit nbrecv preserves the
+	// code's old semantics, distinguishing a sent 0 from "no data".
+	// That said, this code seems to return -1 unconditionally:
+	// the c->keywait channel is unbuffered, and the only sending
+	// to it is done with an nbsendul, which won't block waiting for
+	// a receiver. So there is no sender for nbrecv to find here.
+	if(nbrecv(c->keywait, &u) < 0)
+		return -1;
+	return u;
 }
 
 int
 badkey(Conv *c, Key *k, char *msg, Attr *a)
 {
+	ulong u;
+
 	if(c == nil || *needkeyinuse == 0)
 		return -1;
 
@@ -146,5 +162,17 @@ badkey(Conv *c, Key *k, char *msg, Attr *a)
 		c->tag, k->attr, k->privattr, msg, a);
 	flog("badkey %A / %N / %s / %A",
 		k->attr, k->privattr, msg, a);
-	return nbrecvul(c->keywait);
+
+	// Note: This code used to "return nbrecvul(c->keywait)."
+	// In Jan 2020 we changed nbrecvul to match Plan 9 and
+	// the man page and return 0 on "no data available" instead
+	// of -1. This new code with an explicit nbrecv preserves the
+	// code's old semantics, distinguishing a sent 0 from "no data".
+	// That said, this code seems to return -1 unconditionally:
+	// the c->keywait channel is unbuffered, and the only sending
+	// to it is done with an nbsendul, which won't block waiting for
+	// a receiver. So there is no sender for nbrecv to find here.
+	if(nbrecv(c->keywait, &u) < 0)
+		return -1;
+	return u;
 }
-- 
cgit v1.2.3