Bug 1444228

Created attachment 1273092 [details]
Test script for the bug.

Description of problem:

The GlusterFS build system makes use of the autoconf functionality of generating installed versions of certain text files from templates with .in extensions. In some cases it occurs that autotools managed variables make their ways into the template expansions. It can be a problem (but not always). For example, the mount.glusterfs script is produced as the template expansion of mount.glusterfs.in, which contains:

    prefix="@prefix@";
    exec_prefix=@exec_prefix@;
    cmd_line=$(echo "@sbindir@/glusterfs");

that might expand to (depending on configuration):

    prefix="/usr";
    exec_prefix=${prefix};
    cmd_line=$(echo "${exec_prefix}/sbin/glusterfs");

Here the definitions of exec_prefix and cmd_line include variables but it's not a problem, as mount.glusterfs is a shell script and according to shell script syntax, the variables will be substituted with the actual string values upon running the script.

However, there are other cases, like that of glusterd.vol.in / glusterd.vol, 

    option working-directory @GLUSTERD_WORKDIR@

becaming

    option working-directory ${prefix}/var/lib/glusterd

or that of geo-replication/syncdaemon/conf.py.in / /usr/libexec/glusterfs/glusterfind/conf.py,

    GLUSTERFS_LIBEXECDIR = '@GLUSTERFS_LIBEXECDIR@'
    GLUSTERD_WORKDIR = "@GLUSTERD_WORKDIR@"
    
    LOCALSTATEDIR = "@localstatedir@"
    UUID_FILE = "@GLUSTERD_WORKDIR@/glusterd.info"

becoming

    GLUSTERFS_LIBEXECDIR = '${exec_prefix}/libexec/glusterfs'
    GLUSTERD_WORKDIR = "${prefix}/var/lib/glusterd"
    
    LOCALSTATEDIR = "${prefix}/var"
    UUID_FILE = "${prefix}/var/lib/glusterd/glusterd.info"

where the expanded version of the file functions as config file or python script, which does not support shell style variables and these values will be used verbatim.

After some investigation, I identified the following list of *.in files where variables in the expansion are problematic:

- events/src/eventsapiconf.py.in
- extras/geo-rep/schedule_georep.py.in
- extras/glusterd.vol.in
- extras/snap_scheduler/conf.py.in
- geo-replication/src/peer_georep-sshkey.py.in
- geo-replication/src/peer_mountbroker.in
- geo-replication/src/peer_mountbroker.py.in
- geo-replication/syncdaemon/conf.py.in
- tools/glusterfind/glusterfind.in
- tools/glusterfind/src/tool.conf.in

I've done testing with the following configure invocations:

- ./configure
- ./configure --prefix=/usr
- RHEL's configure invocation
- Ubuntu's configure invocation


How reproducible:

Deterministically.


Steps to Reproduce:

In a GlusterFS source tree with configure script present (ie. either shipped by vendor or ./autogen.sh has been run) run the attached glusterdotin.sh script.

The script calls configure with various arguments (according to the above invocation list). For each invocation it shows information about the invocation (the arguments and summary of results) and a optionally a diff excerpt from the diffs of the possibly affected *.in files and their expanded versions. The diff excerpt shall contain those lines which exhibit the issue.


Actual results:

With some configure invocations a diff excerpt is shown.


Expected results:

No diff excerpt should be shown.


Additional info:

Before change https://review.gluster.org/16880 the issue was present, but among the tested configurations it occurred only with "./configure --prefix=/usr", whereby the following files were affected:
 
- events/src/eventsapiconf.py.in
- geo-replication/src/peer_mountbroker.py.in

However, https://review.gluster.org/16880 worsened the situation to a great extent -- after this change only RHEL's configuration featured flawless template expansion, and for the other three configurations a higher number of files (5 to 9) were affected. (Implied: https://review.gluster.org/16880 is not the final fix for Ubuntu build issues that are reported in the assiciated bug, Bug 1430841.)