[OE-core] [PATCH v3] package.bbclass: allow using EXCLUDE_FROM_SHLIBS for subpackages

[Andrii Bordunov]

Ping

Rebased patch to latest master.

Regarding special value for the variable, it could be done with 
PRIVATE_LIBS to mean all libraries, I think. PRIVATE_LIBS could be set 
per-package already.
Would it be a better option?
However, EXCLUDE_FROM_SHLIBS seems to be a more proper thing to use.


Thank you,
Andrii

On 03/16/2017 05:07 PM, Andrii Bordunov wrote:
> Hi Richard,
> 
> Sorry, my Python knowledge is quite basic. I'm sure it's possible to 
> have a (global?) EXCLUDE_FROM_SHLIBS list holding subpackage names, but 
> how it could lead to performance improvement? It would also need some 
> check for every subpackage.
> Could you please give more details?
> And what do you mean under "all packages"? The intent is like opposite, 
> to split package operations, not to gather.
> 
> As for False parameter to getVar, OK, I just used existing 
> EXCLUDE_FROM_SHLIBS example (seems fixed now).
> 
> In the meantime tried to make a quick and dirty profiling like following:
> 
>    bitbake -c package glibc
>    bitbake -c clean glibc
>    python -m cProfile -s cumtime /mnt/src/oe/distro/bitbake/bin/bitbake 
> -c package glibc > result.txt
> 
> Result is below. If got it right, there are 3833 getVar calls with total 
> time 0.025 second (including subcalls).
> This gives ~6.5 microsecond duration of one call (incl subcalls). Not a 
> lot.
> This is only a partial rebuild of one package of course.
> Also tried to do all of this with and without my patch, but getVar calls 
> number stays exactly the same (w/o modifying glibc recipe, but anyway).
> 
>    NOTE: Tasks Summary: Attempted 314 tasks of which 314 didn't need to 
> be rerun and all succeeded.
>             672869 function calls (667640 primitive calls) in 12.912 
> seconds
>       Ordered by: cumulative time
>       ncalls  tottime  percall  cumtime  percall filename:lineno(function)
>            1    0.001    0.001   12.913   12.913 bitbake:25(<module>)
>            1    0.001    0.001   12.082   12.082 main.py:320(bitbake_main)
>            1    0.005    0.005   11.494   11.494 knotty.py:253(main)
>          250   11.476    0.046   11.476    0.046 {method 'poll' of 
> '_multiprocessing.Connection' objects}
>    ...
>    4481/4046    0.007    0.000    0.026    0.000 
> data_smart.py:666(getVarFlag)
>    ...
>    3833/3719    0.002    0.000    0.025    0.000 data_smart.py:569(getVar)
>    ...
>         1338    0.002    0.000    0.003    0.000 
> data_smart.py:805(getVarFlags)
> 
> 
> 
> Thank you,
> Andrii
> 
> On 16.11.16 12:47, Richard Purdie wrote:
>> On Mon, 2016-10-10 at 20:02 +0300, Andrii Bordunov wrote:
>>> Some packages containing shared libraries might be registered
>>> as shlib providers when they shouldn't (for example, the lib is for
>>> their private use and must not generate any dependency).
>>>
>>> EXCLUDE_FROM_SHLIBS is targeted at that, but it could be set
>>> for entire recipe only.
>>>
>>> This patch expands EXCLUDE_FROM_SHLIBS usage, so now it's possible
>>> to set it in a style similar with RDEPENDS. For example:
>>>  EXCLUDE_FROM_SHLIBS_${PN}-ptest = "1"
>>>
>>> Signed-off-by: Andrii Bordunov <aborduno at cisco.com>
>>> ---
>>>  meta/classes/package.bbclass | 12 ++++++++++--
>>>  1 file changed, 10 insertions(+), 2 deletions(-)
>>
>> My main concern here is actually performance. For a package with many
>> sub packages you just added many calls to getVar when that getVar is
>> extremely unlikely to find any value. Whilst not hugely slow, that
>> operation isn't trivial and hard to fix later.
>>
>> I'm wondering if we could support the syntax
>>
>> EXCLUDE_FROM_SHLIBS = "${PN}-ptest"
>>
>> and then have "1", used in a couple of places as the special value to
>> mean all packages?
>>
>> Also please use False, not 0 as the parameter to getVar.
>>
>> Cheers,
>>
>> Richard

--

 From 4a33d384a72a351bad5a98e46a7c9c1c9755635e Mon Sep 17 00:00:00 2001
From: Andrii Bordunov <aborduno at cisco.com>
Date: Mon, 12 Mar 2018 08:13:44 -0700
Subject: [PATCH v3] Allow EXCLUDE_FROM_SHLIBS to be specified for 
subpackages

Some subpackages containing shared libraries might be registered
as shlib providers when they shouldn't (the lib is for their private use
and must not generate any dependency).

EXCLUDE_FROM_SHLIBS is targeted at that, but it could be set
for entire recipe only.

This patch expands EXCLUDE_FROM_SHLIBS usage, so now it's possible
to set it like following:
EXCLUDE_FROM_SHLIBS_${PN}-ptest = "1"

Signed-off-by: Andrii Bordunov <aborduno at cisco.com>
---
  meta/classes/package.bbclass | 11 +++++++++--
  1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/meta/classes/package.bbclass b/meta/classes/package.bbclass
index 317c775..5d49d41 100644
--- a/meta/classes/package.bbclass
+++ b/meta/classes/package.bbclass
@@ -1571,6 +1571,13 @@ python package_do_shlibs() {

      workdir = d.getVar('WORKDIR')

+    shlib_pkgs = []
+    for pkg in packages.split():
+        if d.getVar('EXCLUDE_FROM_SHLIBS_' + pkg, False):
+            bb.note("not generating shlibs for %s" % pkg)
+        else:
+            shlib_pkgs.append(pkg)
+
      ver = d.getVar('PKGV')
      if not ver:
          msg = "PKGV not defined"
@@ -1698,7 +1705,7 @@ python package_do_shlibs() {
      needed = {}
      shlib_provider = oe.package.read_shlib_providers(d)

-    for pkg in packages.split():
+    for pkg in shlib_pkgs:
          private_libs = d.getVar('PRIVATE_LIBS_' + pkg) or 
d.getVar('PRIVATE_LIBS') or ""
          private_libs = private_libs.split()
          needs_ldconfig = False
@@ -1770,7 +1777,7 @@ python package_do_shlibs() {

      libsearchpath = [d.getVar('libdir'), d.getVar('base_libdir')]

-    for pkg in packages.split():
+    for pkg in shlib_pkgs:
          bb.debug(2, "calculating shlib requirements for %s" % pkg)

          private_libs = d.getVar('PRIVATE_LIBS_' + pkg) or 
d.getVar('PRIVATE_LIBS') or ""
-- 
2.10.3.dirty