0003052: Enabling httpd output causes MPD to exit

Notes
(0006166) mithi (developer) 2010-10-04 21:20	I recently added IPv6 support and a "bind_to_address" option to the httpd output plugin. As a workaround, you could try to add the following line to the httpd output section: bind_to_address "0.0.0.0" This disables IPv6 supports and listens on any IPv4 address, which was the default prior to the change. I'm quite surprised though that you get this error message. "::" is the default value if you have no bind_to_address specified, if at compile-time you had IPv6 support. What's baffling is that your Linux obviously supports IPv6 (looking a the addresses of the ethernet interfaces) but refuses to resolve "::" correctly.

(0006167) cirrus (administrator) 2010-10-04 21:25	mithi, this is not surprising at all. Having IPv6 support at compile time does not mean that it's available at runtime. We pass the option AI_ADDRCONFIG which makes getaddrinfo() care for what's really available. Using an IPv6 address like "::" will lead to an error if IPv6 is not available. You better reuse the code from listen.c, instead of rolling your own. It is very hard to get that right, and one should better use known-good solutions.

(0006168) Brianetta (reporter) 2010-10-04 21:57	The work-around suggested does work.

(0006169) mithi (developer) 2010-10-04 22:27 edited on: 2010-10-04 22:29	There is no code in listen.c that does what httpd needs. The main difference is that httpd doesn't immediately create a socket but delays opening the socket until the plugin is opened, which in particular means that we can't reuse the listen_add_port*-functions. I agree that it's bad behavior to break existing configs (without a specified bind_to_address). It's basically just a question of what the default should be. On the one hand we could argue that since IPv4-only was the previous default, we should stick to IPv4-only with unconfigured bind_to_address. On the other hand we could argue that the default should be to support as much as possible, which would be IPv6 if available. But that raises the question of what to do on dual-stack systems, where "::" doesn't include IPv4-mapped addresses. This is however not my call to make. The former would be a really simple patch, the latter would require a fallback mechanism which shouldn't be so hard to devise. Edit: This however doesn't explain why IPv6 isn't available even though the interfaces clearly have IPv6 link-local addresses assigned.

(0006170) cirrus (administrator) 2010-10-04 22:39	listen.c is not generic enough, but it can be made so. This was the main reason I hadn't implemented that patch yet; I wanted to clean up listen.c first. If you reimplement it instead (without learning from it), you are doomed to repeat the same mistakes listen.c had in its early days.

(0006171) mithi (developer) 2010-10-04 23:03	Then please revert my patch, so it won't cause trouble for other users.

Issue History
Date Modified	Username	Field	Change
2010-10-04 00:25	Brianetta	New Issue
2010-10-04 00:25	Brianetta	Status	new => assigned
2010-10-04 00:25	Brianetta	Assigned To	=> cirrus
2010-10-04 00:25	Brianetta	File Added: mpd.conf
2010-10-04 21:20	mithi	Note Added: 0006166
2010-10-04 21:25	cirrus	Note Added: 0006167
2010-10-04 21:57	Brianetta	Note Added: 0006168
2010-10-04 22:27	mithi	Note Added: 0006169
2010-10-04 22:29	mithi	Note Edited: 0006169	View Revisions
2010-10-04 22:39	cirrus	Note Added: 0006170
2010-10-04 23:03	mithi	Note Added: 0006171
2010-10-05 21:24	cirrus	Status	assigned => resolved
2010-10-05 21:24	cirrus	Fixed in Version	=> git
2010-10-05 21:24	cirrus	Resolution	open => fixed
2010-10-06 12:56	cirrus	Relationship added	has duplicate 0003057
2011-03-19 21:22	Avuton Olrich	Status	resolved => closed