icedtea7-native build uses free -m (from the procfs suite of tools) to
determine memory bound during compile. This always fails for us,
because free is not passed through from the host environment.
This results in a few warnings in the logs:
/bin/sh: line 1: free: command not found
[..snip..]/Sanity-Settings.gmk:195: WARNING: MB_OF_MEMORY should not
be empty [Sanity-Settings.gmk]
But otherwise this is handled gracefully. Problems arise when building
icedtea7-native on hosts with more modest RAM sizes: Build process
will consume so much RAM that the OOM killer is invoked. Let us avoid
this by passing through free from the host environment.
We use HOSTTOOLS_NONFATAL instead of HOSTTOOLS as to not break
currently working setups that have enough RAM but lack free(1).
This needs to be done globally due to the way HOSTTOOLS_NONFATAL is
evaluated, we thus stick it into the layer.conf.
Signed-off-by: Ahmad Fatoum <[email protected]>
---
If patch is deemed acceptable, can this be backported to kirkstone too?
Thanks.
---
conf/layer.conf | 2 ++
1 file changed, 2 insertions(+)
diff --git a/conf/layer.conf b/conf/layer.conf
index 2e2ce1677ecc..bd8f5aebaacb 100644
--- a/conf/layer.conf
+++ b/conf/layer.conf
@@ -12,6 +12,8 @@ LAYERDEPENDS_meta-java += "openembedded-layer"
LICENSE_PATH += "${LAYERDIR}/licenses"
+HOSTTOOLS_NONFATAL += "free"
+
SIGGEN_EXCLUDERECIPES_ABISAFE += "\
ca-certificates-java \
"
--
2.30.2
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#99338):
https://lists.openembedded.org/g/openembedded-devel/message/99338
Mute This Topic: https://lists.openembedded.org/mt/94630914/21656
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-devel/unsub
[[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-
