Re: [OE-core] [meta-oe][PATCH v2] procps: split into binary subpackages

["Richard Purdie" <richard.purdie@linuxfoundation.org>]

On Tue, 2020-12-01 at 03:43 +0000, Sinan Kaya wrote:
> This change allows you to pull individual binaries into the
> target without including the entire suite.
> 
> Signed-off-by: Sinan Kaya <okaya@kernel.org>
> ---
>  meta/recipes-extended/procps/procps_3.3.16.bb | 51 +++++++++++++++++--
>  1 file changed, 48 insertions(+), 3 deletions(-)
> 
> diff --git a/meta/recipes-extended/procps/procps_3.3.16.bb b/meta/recipes-extended/procps/procps_3.3.16.bb
> index 2810ebd285a..1b62ae33f5c 100644
> --- a/meta/recipes-extended/procps/procps_3.3.16.bb
> +++ b/meta/recipes-extended/procps/procps_3.3.16.bb
> @@ -50,15 +50,13 @@ do_install_append () {
>  
>  CONFFILES_${PN} = "${sysconfdir}/sysctl.conf"
>  
> -bindir_progs = "free pkill pmap pgrep pwdx skill snice top uptime w"
> +bindir_progs = "free pkill pmap pgrep pwdx skill slabtop snice tload top uptime vmstat w"
>  base_bindir_progs += "kill pidof ps watch"
>  base_sbindir_progs += "sysctl"
>  
>  ALTERNATIVE_PRIORITY = "200"
>  ALTERNATIVE_PRIORITY[pidof] = "150"
>  
> -ALTERNATIVE_${PN} = "${bindir_progs} ${base_bindir_progs} ${base_sbindir_progs}"
> -
>  ALTERNATIVE_${PN}-doc = "kill.1 uptime.1"
>  ALTERNATIVE_LINK_NAME[kill.1] = "${mandir}/man1/kill.1"
>  ALTERNATIVE_LINK_NAME[uptime.1] = "${mandir}/man1/uptime.1"
> @@ -69,8 +67,55 @@ python __anonymous() {
>  
>      for prog in d.getVar('base_sbindir_progs').split():
>          d.setVarFlag('ALTERNATIVE_LINK_NAME', prog, '%s/%s' % (d.getVar('base_sbindir'), prog))
> +
> +    for prog in d.getVar('bindir_progs').split():
> +        d.setVarFlag('ALTERNATIVE_LINK_NAME', prog, '%s/%s' % (d.getVar('bindir'), prog))
>  }
>  
>  # 'ps' isn't suitable for use as a security tool so whitelist this CVE.
>  # https://bugzilla.redhat.com/show_bug.cgi?id=1575473#c3
>  CVE_CHECK_WHITELIST += "CVE-2018-1121"
> +
> +python __anonymous () {
> +    packages = []
> +    for prog in d.getVar('bindir_progs').split():
> +        pkg = "procps-%s" % prog.replace("_", "-")
> +        packages.append(pkg)
> +        d.setVar("RDEPENDS_%s" % pkg, "procps-lib")
> +        d.setVar("ALTERNATIVE_%s" % pkg, prog.replace("_", "-"))
> +        if not d.getVar("FILES_%s" % pkg):
> +                d.setVar("FILES_%s" % pkg, "${bindir}/%s*" % prog)
> +    for prog in d.getVar('base_sbindir_progs').split():
> +        pkg = "procps-%s" % prog.replace("_", "-")
> +        packages.append(pkg)
> +        d.setVar("RDEPENDS_%s" % pkg, "procps-lib")
> +        d.setVar("ALTERNATIVE_%s" % pkg, prog.replace("_", "-"))
> +        if not d.getVar("FILES_%s" % pkg):
> +                d.setVar("FILES_%s" % pkg, "${base_sbindir}/%s*" % prog)
> +    for prog in d.getVar('base_bindir_progs').split():
> +        pkg = "procps-%s" % prog.replace("_", "-")
> +        packages.append(pkg)
> +        d.setVar("RDEPENDS_%s" % pkg, "procps-lib")
> +        d.setVar("ALTERNATIVE_%s" % pkg, prog.replace("_", "-"))
> +        if not d.getVar("FILES_%s" % pkg):
> +                d.setVar("FILES_%s" % pkg, "${base_bindir}/%s*" % prog)
> +    pkg = "procps-man"
> +    packages.append(pkg)
> +    if not d.getVar("FILES_%s" % pkg):
> +            d.setVar("FILES_%s" % pkg, "${mandir}/*")
> +    pkg = "procps-lib"
> +    packages.append(pkg)
> +    if not d.getVar("FILES_%s" % pkg):
> +            d.setVar("FILES_%s" % pkg, "${libdir}/*")
> +    pkg = "procps-sysctl"
> +    d.appendVar("FILES_%s" % pkg, " ${sysconfdir}/sysctl.conf")
> +
> +    d.setVar("PROCPS_PACKAGES", " ".join(packages))
> +}
> +# "procps" is a metapackage which pulls in all procps binaries
> +PACKAGES += "${PROCPS_PACKAGES}"
> +PROVIDES += "${PROCPS_PACKAGES}"
> +FILES_${PN} = ""
> +ALLOW_EMPTY_${PN} = "1"
> +RRECOMMENDS_${PN} += "${PROCPS_PACKAGES}"

I am starting to get a little worried about the direction these patches
are heading in. How much of the system are we going to split into
individual package per binaries?

At some point the packages stop becoming useful and you have to
consider whather package management is the best way to build these
rootfs since I doubt you're enabling package management in the end
result anyway given the size gains you're aiming for?

I'd like to understand and discuss the end goal here before we merge
too many large chunks of custom python like the above?

An alternative may be to create a library function which can be
optionally enabled for recipes as this code and the other code you sent
does look similar, at least superficially...

Cheers,

Richard