gawk: backport randtest stability fix
PATCH 1/1 gawk: backport randtest stability fix
Backport upstream fix to improve stability of randtest.sh by retrying when sigma is too small, avoiding intermittent failures.
Upstream-Status: Backport [f2250f4bc864913437619af7a9834a1c24915acd]
--- .../gawk/files/randtest-fix.patch | 138 ++++++++++++++++++ meta/recipes-extended/gawk/gawk_5.3.0.bb | 1 + 2 files changed, 139 insertions(+) create mode 100644 meta/recipes-extended/gawk/files/randtest-fix.patch
meta/recipes-extended/gawk/files/randtest-fix.patch new file +138 -0
@@ -0,0 +1,138 @@ +From f2250f4bc864913437619af7a9834a1c24915acd Mon Sep 17 00:00:00 2001 +From: "Arnold D. Robbins" <[email protected]> +Date: Thu, 28 May 2026 09:52:41 -0400 +Subject: [PATCH] Improve test/randtest.sh. + +Backport upstream fix to improve stability of randtest.sh by retrying +when sigma is too small, avoiding intermittent failures. + +Upstream-Status: Backport [f2250f4bc864913437619af7a9834a1c24915acd] + +--- + +diff --git a/test/randtest.sh b/test/randtest.sh +index 597376a8..6ae6f830 100755 +--- a/test/randtest.sh ++++ b/test/randtest.sh +@@ -45,68 +45,77 @@ $AWK 'BEGIN{ + nsamples=('$NSAMPLES'); + max_allowed_sigma=('$MAX_ALLOWED_SIGMA'); + nruns=('$NRUNS'); +- for(tau=0;tau<nsamples/2;tau++) corr[tau]=0; ++ max_retries=5 + +- for(run=0;run<nruns;run++) { +- sum=0; ++ for (retry=0; retry<max_retries; retry++) { ++ for(tau=0; tau<nsamples/2; tau++) corr[tau]=0; + +- # Fill an array with a sequence of samples that are a +- # function of pairs of rand() values. ++ for(run=0; run<nruns; run++) { ++ sum=0; + +- for(i=0;i<nsamples;i++) { +- samp[i]=((rand()-0.5)*(rand()-0.5))^2; +- sum=sum+samp[i]; +- } ++ # Fill an array with a sequence of samples that are a ++ # function of pairs of rand() values. + +- # Subtract off the mean of the sequence: ++ for(i=0; i<nsamples; i++) { ++ samp[i]=((rand()-0.5)*(rand()-0.5))^2; ++ sum=sum+samp[i]; ++ } + +- mean=sum/nsamples; +- for(i=0;i<nsamples;i++) samp[i]=samp[i]-mean; ++ # Subtract off the mean of the sequence: + +- # Calculate an autocorrelation function on the sequence. +- # Because the values of rand() should be independent, there +- # should be no peaks in the autocorrelation. ++ mean=sum/nsamples; ++ for(i=0;i<nsamples;i++) samp[i]=samp[i]-mean; + +- for(tau=0;tau<nsamples/2;tau++) { +- sum=0; +- for(i=0;i<nsamples/2;i++) sum=sum+samp[i]*samp[i+tau]; +- corr[tau]=corr[tau]+sum; +- } ++ # Calculate an autocorrelation function on the sequence. ++ # Because the values of rand() should be independent, there ++ # should be no peaks in the autocorrelation. + +- } +- # Normalize the autocorrelation to the tau=0 value. ++ for(tau=0; tau<nsamples/2; tau++) { ++ sum=0; ++ for(i=0;i<nsamples/2;i++) sum=sum+samp[i]*samp[i+tau]; ++ corr[tau]=corr[tau]+sum; ++ } + +- max_corr=corr[0]; +- for(tau=0;tau<nsamples/2;tau++) corr[tau]=corr[tau]/max_corr; ++ } ++ # Normalize the autocorrelation to the tau=0 value. + +- # OPTIONALLY Print out the autocorrelation values: ++ max_corr=corr[0]; ++ for(tau=0;tau<nsamples/2;tau++) corr[tau]=corr[tau]/max_corr; + +- # for(tau=0;tau<nsamples/2;tau++) print tau, corr[tau] > "pairpower_corr.data"; ++ # OPTIONALLY Print out the autocorrelation values: + +- # Calculate the sigma for the non-zero tau values: ++ # for(tau=0;tau<nsamples/2;tau++) print tau, corr[tau] > "pairpower_corr.data"; + +- power_sum=0; ++ # Calculate the sigma for the non-zero tau values: + +- for(tau=1;tau<nsamples/2;tau++) power_sum=power_sum+(corr[tau])^2; ++ power_sum=0; + +- sigma=sqrt(power_sum/(nsamples/2-1)); ++ for(tau=1;tau<nsamples/2;tau++) power_sum=power_sum+(corr[tau])^2; + +- # See if any of the correlations exceed a reasonable number of sigma: ++ sigma=sqrt(power_sum/(nsamples/2-1)); + +- passed=1; +- for(tau=1;tau<nsamples/2;tau++) { +- if ( abs(corr[tau])/sigma > max_allowed_sigma ) { +- print "Tau=", tau ", Autocorr=", corr[tau]/sigma, "sigma"; +- passed=0; +- } ++ if (sigma < 1e-6) { ++ print "Sigma too small, retrying run", retry ++ continue + } +- if(!passed) { +- print "Test failed." +- exit(1); ++ ++ # See if any of the correlations exceed a reasonable number of sigma: ++ ++ for(tau=1;tau<nsamples/2;tau++) { ++ if (abs(corr[tau])/sigma > max_allowed_sigma) { ++ print "Tau=", tau ", Autocorr=", corr[tau]/sigma, "sigma"; ++ print "Test failed." ++ exit 1 ++ } + } +- else exit (0); ++ ++ exit 0 + } + ++ print "Test failed after retries due to unstable sigma" ++ exit 1 ++} ++ + function abs(abs_input) { return(sqrt(abs_input^2)) ; } + ' + +-- +2.34.1 +
meta/recipes-extended/gawk/gawk_5.3.0.bb +1 -0
@@ -19,6 +19,7 @@ SRC_URI = "${GNU_MIRROR}/gawk/gawk-${PV}.tar.gz \ file://0001-m4-readline-add-missing-includes.patch \ file://run-ptest \ " +SRC_URI += "file://randtest-fix.patch" SRC_URI[sha256sum] = "378f8864ec21cfceaa048f7e1869ac9b4597b449087caf1eb55e440d30273336"
Signature · 1 line
2.34.1
Quoted 5 lines · Aditya GS
Backport upstream fix to improve stability of
randtest.shby retrying when sigma is too small, avoiding intermittent failures.Upstream-Status: Backport [
f2250f4bc864913437619af7a9834a1c24915acd]
Upstream-Status should only be in the patch file itself, not in the commit message for oe-core.
Instead, provide a reference to the yocto bugzilla, so that the reviewers know which bug this is fixing. A http link or [YOCTO #nnnnnn] tag.
Quoted 1 line
+Upstream-Status: Backport [
f2250f4bc864913437619af7a9834a1c24915acd]
It's better to link to the web interface:
Upstream-Status: Backport
[https://cgit.git.savannah.gnu.org/cgit/gawk.git/commit/?id=f2250f4bc864913437619af7a9834a1c24915acd]
The reason is that sometimes a patch submission is labeled as Backport, but actually isn't (not here, of course). It's best to make that easy to verify, with a link like above.
meta/recipes-extended/gawk/gawk_5.3.0.bb SRC_URI = "${GNU_MIRROR}/gawk/gawk-${PV}.tar.gz \ · line 22
Quoted 5 lines
meta/recipes-extended/gawk/gawk_5.3.0.bb +1
-0@@ -19,6 +19,7 @@ SRC_URI = "${GNU_MIRROR}/gawk/gawk-${PV}.tar.gz \ file://0001-m4-readline-add-missing-includes.patch \ file://run-ptest \ " +SRC_URI += "file://randtest-fix.patch"
You can simply change the main SRC_URI assignment and add the patch there, instead of making a new statement with +=.
Other than this, looks good. Thanks for taking time to develop a proper fix, and getting upstream to accept it. It's much better to solve the issue properly, than simply skip the flaky tests.
Alex
Hi Alex,
Thanks for the detailed review and suggestions.
I've updated the patch based on your feedback:
Removed Upstream-Status from commit message and added Bugzilla reference
Updated Upstream-Status in the patch with the upstream commit link
Integrated the patch into the main SRC_URI
The updated patch has now been sent to the mailing list for review.
Please let me know if any further changes are needed.
Thanks,
Aditya
Quoted 42 lines · Alexander Kanavin
Quoted 5 lines · Aditya GS
Backport upstream fix to improve stability of
randtest.shby retrying when sigma is too small, avoiding intermittent failures.Upstream-Status: Backport [
f2250f4bc864913437619af7a9834a1c24915acd]Upstream-Status should only be in the patch file itself, not in the commit message for oe-core.
Instead, provide a reference to the yocto bugzilla, so that the reviewers know which bug this is fixing. A http link or [YOCTO #nnnnnn] tag.
Quoted 1 line
+Upstream-Status: Backport [
f2250f4bc864913437619af7a9834a1c24915acd]It's better to link to the web interface:
Upstream-Status: Backport
[
https://cgit.git.savannah.gnu.org/cgit/gawk.git/commit/?id=f2250f4bc864913437619af7a9834a1c24915acd
]The reason is that sometimes a patch submission is labeled as Backport, but actually isn't (not here, of course). It's best to make that easy to verify, with a link like above.
meta/recipes-extended/gawk/gawk_5.3.0.bb SRC_URI = "${GNU_MIRROR}/gawk/gawk-${PV}.tar.gz \ · line 22
Quoted 5 lines
meta/recipes-extended/gawk/gawk_5.3.0.bb +1
-0@@ -19,6 +19,7 @@ SRC_URI = "${GNU_MIRROR}/gawk/gawk-${PV}.tar.gz \ file://0001-m4-readline-add-missing-includes.patch \ file://run-ptest \ " +SRC_URI += "file://randtest-fix.patch"You can simply change the main
SRC_URIassignment and add the patch there, instead of making a new statement with +=.Other than this, looks good. Thanks for taking time to develop a proper fix, and getting upstream to accept it. It's much better to solve the issue properly, than simply skip the flaky tests.
Alex