Project

General

Profile

Bug #6771

end-of-loop code not reached in common/dnssd_clientstub.c

Added by Gary Mills almost 5 years ago. Updated almost 5 years ago.

Status:
Closed
Priority:
Normal
Assignee:
Gary Mills
Category:
lib - userland libraries
Start date:
2016-03-20
Due date:
% Done:

100%

Estimated time:
Difficulty:
Bite-size
Tags:
needs-triage
Gerrit CR:

Description

A build with Studio as the shadow compiler produced this error:

"../common/dnssd_clientstub.c", line 675: end-of-loop code not reached (E_END_OF_LOOP_CODE_NOT_REACHED)
cc: acomp failed for ../common/dnssd_clientstub.c

Here's the code in usr/src/lib/libdns_sd/common/dnssd_clientstub.c:

    599 static DNSServiceErrorType deliver_request(ipc_msg_hdr *hdr, DNSServiceOp *sdr)
    600 {
    ...
    633         #if defined(USE_TCP_LOOPBACK)
    ...
    651         #elif defined(USE_NAMED_ERROR_RETURN_SOCKET)
    ...
    672         #else
    673         {
    674             dnssd_sock_t sp[2];
-------------------------------------------------------------------------------
    675             if (socketpair(AF_DNSSD, SOCK_STREAM, 0, sp) < 0) deliver_request_bailout("socketpair");
-------------------------------------------------------------------------------
    676             else
    677             {
    ...
    689         #endif

The deliver_request_bailout macro is defined this way:

+ #define deliver_request_bailout(MSG) \
      do { syslog(LOG_WARNING, "dnssd_clientstub deliver_request: %s failed %d (%s)", (MSG), dnssd_errno, dnssd_strerror(dnssd_errno)); goto cleanup; } while(0)

The compiler is complaining about the jump out of the do/while loop. There are two possible solutions. One is to alter the macro by removing the loop. The other is to suppress the error by adding -erroff=E_END_OF_LOOP_CODE_NOT_REACHED to the appropriate Makefile.

#1

Updated by Toomas Soome almost 5 years ago

Gary Mills wrote:

A build with Studio as the shadow compiler produced this error:

[...]

Here's the code in usr/src/lib/libdns_sd/common/dnssd_clientstub.c:

[...]

The deliver_request_bailout macro is defined this way:

[...]

The compiler is complaining about the jump out of the do/while loop. There are two possible solutions. One is to alter the macro by removing the loop. The other is to suppress the error by adding -erroff=E_END_OF_LOOP_CODE_NOT_REACHED to the appropriate Makefile.

Hm, are your build options somewhat different or was this appearing on sparc? Because this code was specifically cleaned/tested with shadow build enabled and I'm curious why this case was not picked up back then. I could run shadow build now again to see if it is still clean or not...

#2

Updated by Gary Mills almost 5 years ago

This build was done on SPARC, but that shouldn't have made any difference in this case. I suspect the compiler. It was:

64-bit compiler
/opt/onbld/bin/sparc/cw -_gcc
cw version 1.30
primary: /opt/gcc/4.4.4/bin/gcc
gcc (Illumos tags/gcc-4.4.4-il-2) 4.4.4
shadow: /opt/SUNWspro/bin/cc
cc: Sun C 5.10 SunOS_sparc 2009/06/03

gcc did compile it without errors.

#3

Updated by Toomas Soome almost 5 years ago

Gary Mills wrote:

This build was done on SPARC, but that shouldn't have made any difference in this case. I suspect the compiler. It was:

[...]

gcc did compile it without errors.

yep, gcc was ok on first attempt and on pre-integration build Robert did discover the shadow compile fail. on x86 the cc is the same:
cc: Sun C 5.10 SunOS_i386 2009/06/03

so I suspect the cc does behave differently on x86 and sparc, which explains the error. To be honest, I don't have preference about which way to go - to pollute Makefile with cc flags or remove do while() - to clean up the cc errors. I had to update the source anyhow, so adding this one does not make much difference ...

#4

Updated by Gary Mills almost 5 years ago

Adding a line to Makefile.com proved be the simplest change. I'll also test this change on x86, just to ensure that it doesn't cause any problems there. I'll be submitting a webrev soon.

#5

Updated by Gary Mills almost 5 years ago

Well, the simple change seemed to be unacceptable by the advocates. I'm doing the more complicated one now, with changes to the definition and invocations of the deliver_request_bailout macro in dnssd_clientstub.c . The build is also successful, both on SPARC and x86. I'll be submitting a webrev soon.

#6

Updated by Electric Monk almost 5 years ago

  • Status changed from Feedback to Closed
  • % Done changed from 40 to 100

git commit 5dee1fa64fd5ddfe4221aaa94bc981e0b9d78b92

commit  5dee1fa64fd5ddfe4221aaa94bc981e0b9d78b92
Author: Gary Mills <gary_mills@fastmail.fm>
Date:   2016-04-04T21:19:37.000Z

    6771 end-of-loop code not reached in common/dnssd_clientstub.c
    Reviewed by: Toomas Soome <tsoome@me.com>
    Reviewed by: "Garrett D'Amore" <garrett@damore.org>
    Approved by: Robert Mustacchi <rm@joyent.com>

Also available in: Atom PDF