On 11/07/2014 06:42 AM, Burton, Ross wrote:

On 4 November 2014 05:04, Qian Lei <qianl.fnst@cn.fujitsu.com> wrote:
        # at_console doesn't really work with the current state of OE, so punch some more holes so people can actually use BT
+        install -m 0755 ${S}/attrib/gatttool ${D}/${bindir}/
        install -m 0644 ${WORKDIR}/bluetooth.conf ${D}/${sysconfdir}/dbus-1/system.d/
 }

Because of where you put the install, it looks like installing gatttool is related to the DBus configuration.
 
 ALLOW_EMPTY_libasound-module-bluez = "1"
-PACKAGES =+ "libasound-module-bluez ${PN}-testtools ${PN}-obex"
+PACKAGES =+ "libasound-module-bluez ${PN}-testtools ${PN}-obex ${PN}-tools"

Two questions:
1) If we're installing gatttool, why not the other tools that are noinst (obex-client-tool, obexctl, etc).  Patching the makefile would install all the tools that upstream build.
2) Do these deserve a separate package, or as they're for testing should they go into PN-testtools? 


I do think the extra tools deserve an extra package, because they may or may not be for testing, and in any case don't get installed in a distinct test area as the current contents of ${PN}-testtools do.

In fact, they don't get installed at all.  If EXPERIMENTAL is enabled via PACKAGECONFIG (as it is in one of my WIP virtual/bluez patches) there are a lot of those tools, some of which have opaque names that don't appear related to bluetooth.  However, as I use bluez5 more of them are things I want available on the target.

At this time I'm inclined to add a ${PN}-noinst-tools package for them.  I'm debating whether it's better to manually install them from the build area, to highlight that upstream doesn't install them, or to modify the Makefile to remove "noinst_" and then explicitly list each one in a separate FILES_${PN}-noinst-tools variable.  The latter seems a higher risk for unwittingly adding non-standard executables to ${PN}.

Is there any precedent for a package to have its own subdirectory off ${bindir}?  It'd be easy enough to patch the Makefile to put them all in ${bindir}/bluez5.

I'm sympathetic with Qian Lei and also assumed that gattool, which used to be installed in bluez4, might be an exception and just belongs as an OE addition to bluez5 rather than being put in ${PN}-noinst-tools.  But that's a slippery slope and new things like btgatt-client might be just as important to the folks who want to use GATT with bluez5.  Now I'm inclined to keep it separate too.

Peter