Skip to contentSkip to conversation

What is TOOLCHAIN_NEED_CONFIGSITE_CACHE for?

DISCUSSION Phil Blundell 4 participants 5 messages Oct 29, 2012

lore
Share & tools
lore
#0 lore raw

This morning I happened to notice that some/all of my images had a dependency on ncurses for no obviously good reason. (Following an earlier small crusade these are now the only remaining dependencies on ncurses in my build, so I would quite like to get rid of them.)

It turns out that the dependency in question is due to the code in toolchain-scripts.bbclass (which is inherited from populate_sdk_base.bbclass, from image.bbclass) which does:

TOOLCHAIN_NEED_CONFIGSITE_CACHE = "${TCLIBC} ncurses"

[...]

python __anonymous () {
deps = ""
for dep in (d.getVar('TOOLCHAIN_NEED_CONFIGSITE_CACHE', True) or "").split():
deps += " %s:do_populate_sysroot" % dep
d.appendVarFlag('do_configure', 'depends', deps)
}

The code in question was added in:

commit 9e87f1347788beed181476dc4563085db14a4729

Author: Lianhao Lu <[email protected]>
Date:   Tue Aug 16 11:15:36 2011 +0800

toolchain-script.bbclass: Collected cached site config in runtime.

[YOCTO #892]
Modify the function toolchain_create_sdk_siteconfig to collect the cached site config files which are specified by TOOLCHAIN_NEED_CONFIGSITE_CACHE in runtime.

Also added task dependency to ensure the cached site config files are generated.

... but I don't entirely understand from that commentary what exactly is special about ncurses that means it needs to be handled this way. Can anybody clarify?

thanks

p.

#1 lore raw
Quoted 33 lines · Phil Blundell

This morning I happened to notice that some/all of my images had a dependency on ncurses for no obviously good reason. (Following an earlier small crusade these are now the only remaining dependencies on ncurses in my build, so I would quite like to get rid of them.)

It turns out that the dependency in question is due to the code in toolchain-scripts.bbclass (which is inherited from populate_sdk_base.bbclass, from image.bbclass) which does:

TOOLCHAIN_NEED_CONFIGSITE_CACHE = "${TCLIBC} ncurses"

[...]

python __anonymous () {
deps = ""
for dep in (d.getVar('TOOLCHAIN_NEED_CONFIGSITE_CACHE', True) or "").split():
deps += " %s:do_populate_sysroot" % dep
d.appendVarFlag('do_configure', 'depends', deps)
}

The code in question was added in:

commit 9e87f1347788beed181476dc4563085db14a4729

Author: Lianhao Lu <[email protected]>
Date:   Tue Aug 16 11:15:36 2011 +0800

toolchain-script.bbclass: Collected cached site config in runtime.

[YOCTO #892]
Modify the function toolchain_create_sdk_siteconfig to collect the cached site config files which are specified by TOOLCHAIN_NEED_CONFIGSITE_CACHE in runtime.

Quoted 8 lines

Also added task dependency to ensure the cached site config files are generated.

... but I don't entirely understand from that commentary what exactly is special about ncurses that means it needs to be handled this way. Can anybody clarify?

We ship the "site" cache files with the toolchain. We decided to include the generated cache files as well as the static ones. We currently generate "site" files for libc and ncurses.

You can almost certainly just set:

TOOLCHAIN_NEED_CONFIGSITE_CACHE = "${TCLIBC}"

and be happy since this just lists which generated cache site files to include.

Cheers,

Richard

#2 lore raw
Quoted 47 lines · Richard Purdie
Quoted 33 lines · Phil Blundell

This morning I happened to notice that some/all of my images had a dependency on ncurses for no obviously good reason. (Following an earlier small crusade these are now the only remaining dependencies on ncurses in my build, so I would quite like to get rid of them.)

It turns out that the dependency in question is due to the code in toolchain-scripts.bbclass (which is inherited from populate_sdk_base.bbclass, from image.bbclass) which does:

TOOLCHAIN_NEED_CONFIGSITE_CACHE = "${TCLIBC} ncurses"

[...]

python __anonymous () {
deps = ""
for dep in (d.getVar('TOOLCHAIN_NEED_CONFIGSITE_CACHE', True) or "").split():
deps += " %s:do_populate_sysroot" % dep
d.appendVarFlag('do_configure', 'depends', deps)
}

The code in question was added in:

commit 9e87f1347788beed181476dc4563085db14a4729

Author: Lianhao Lu <[email protected]>
Date:   Tue Aug 16 11:15:36 2011 +0800

toolchain-script.bbclass: Collected cached site config in runtime.

[YOCTO #892]
Modify the function toolchain_create_sdk_siteconfig to collect the cached site config files which are specified by TOOLCHAIN_NEED_CONFIGSITE_CACHE in runtime.

Quoted 8 lines

Also added task dependency to ensure the cached site config files are generated.

... but I don't entirely understand from that commentary what exactly is special about ncurses that means it needs to be handled this way. Can anybody clarify?

We ship the "site" cache files with the toolchain. We decided to include the generated cache files as well as the static ones. We currently generate "site" files for libc and ncurses.

Quoted 6 lines

You can almost certainly just set:

TOOLCHAIN_NEED_CONFIGSITE_CACHE = "${TCLIBC}"

and be happy since this just lists which generated cache site files to include.

Shouldn't it be pulled in via depends from the populate_sdk bits, rather than a task that's used both for image builds and toolchain builds, though?

Signature · 1 line
Christopher Larson
#3 lore raw
Quoted 51 lines · Chris Larson
Quoted 49 lines · Richard Purdie
Quoted 33 lines · Phil Blundell

This morning I happened to notice that some/all of my images had a dependency on ncurses for no obviously good reason. (Following an earlier small crusade these are now the only remaining dependencies on ncurses in my build, so I would quite like to get rid of them.)

It turns out that the dependency in question is due to the code in toolchain-scripts.bbclass (which is inherited from populate_sdk_base.bbclass, from image.bbclass) which does:

TOOLCHAIN_NEED_CONFIGSITE_CACHE = "${TCLIBC} ncurses"

[...]

python __anonymous () {
deps = ""
for dep in (d.getVar('TOOLCHAIN_NEED_CONFIGSITE_CACHE', True) or "").split():
deps += " %s:do_populate_sysroot" % dep
d.appendVarFlag('do_configure', 'depends', deps)
}

The code in question was added in:

commit 9e87f1347788beed181476dc4563085db14a4729

Author: Lianhao Lu <[email protected]>
Date:   Tue Aug 16 11:15:36 2011 +0800

toolchain-script.bbclass: Collected cached site config in runtime.

[YOCTO #892]
Modify the function toolchain_create_sdk_siteconfig to collect the cached site config files which are specified by TOOLCHAIN_NEED_CONFIGSITE_CACHE in runtime.

Quoted 8 lines

Also added task dependency to ensure the cached site config files are generated.

... but I don't entirely understand from that commentary what exactly is special about ncurses that means it needs to be handled this way. Can anybody clarify?

We ship the "site" cache files with the toolchain. We decided to include the generated cache files as well as the static ones. We currently generate "site" files for libc and ncurses.

You can almost certainly just set:

Quoted 9 lines
Quoted 4 lines

TOOLCHAIN_NEED_CONFIGSITE_CACHE = "${TCLIBC}"

and be happy since this just lists which generated cache site files to include.

Shouldn't it be pulled in via depends from the populate_sdk bits, rather than a task that's used both for image builds and toolchain builds, though?

Ya, this seems like a design issue. The site cache files require the package to be built, but I was under the impression they were included with the dev items within that package. If this isn't how it works, maybe this is a change we should consider.

--Mark

#4 lore raw
In reply to Chris Larson
Quoted 9 lines · Chris Larson
Quoted 7 lines · Richard Purdie

We ship the "site" cache files with the toolchain. We decided to include the generated cache files as well as the static ones. We currently generate "site" files for libc and ncurses.

You can almost certainly just set:

Quoted 8 lines
Quoted 4 lines

TOOLCHAIN_NEED_CONFIGSITE_CACHE = "${TCLIBC}"

and be happy since this just lists which generated cache site files to include.

Shouldn't it be pulled in via depends from the populate_sdk bits, rather than a task that's used both for image builds and toolchain builds, though?

If we turned these into packaged output then we could do this, yes. As things stand those files aren't packaged and hence there are no dependencies from which to infer/pull these in. Converting them to that isn't a particularly trivial exercise.

We currently only generate them for libc/ncurses which provide most of the configure cache speedups we'd want so it hasn't been a particularly pressing problem to solve.

I'm all for improving this and will happily take patches but its not something on the immediate task list as there are higher priority issues IMO.

Cheers,

Richard