[golang-dev] Re: code review 6114046: runtime, reflect, ld, gc: garbage collection precision ... (issue 6114046)

Hi. I am unable to reproduce the issue on linux/amd64, rev 87f67aadaed6.

If it also fails with rev 87f67aadaed6 on your machine: Does the failure
occur randomly, or every time you test encoding/xml?

https://codereview.appspot.com/6114046/

randomly but with a high probability of failure (all about the same
test TestUnmarshalFeed).

for example:
--- FAIL: TestUnmarshalFeed (0.00 seconds)
read_test.go:21: have xml.Feed{XMLName:xml.Name{Space:"
http://www.w3.org/2005/Atom", Local:"feed"}, Title:"Code Review - My
issues", Id:"http://codereview.appspot.com/",
Link:[]xml.Link{xml.Link{Rel:"alternate", Href:"
http://codereview.appspot.com/"}, xml.Link{Rel:"self", Href:"
http://codereview.appspot.com/rss/mine/rsc"}},
Updated:time.Time{sec:63390216958, nsec:0,
loc:(*time.Location)(0xc2000f4180)}, Author:xml.Person{Name:"rietveld<>",
URI:"", Email:"", InnerXML:"<name>rietveld&lt;&gt;</name>"},
Entry:[]xml.Entry{xml.Entry{Title:"rietveld: an attempt at pubsubhubbub\n",
Id:"urn:md5:134d9179c41f806be79b3a5f7877d19a",
Link:[]xml.Link{xml.Link{Rel:"alternate", Href:"
http://codereview.appspot.com/126085"}}, Updated:time.Time{sec:63390216958,
nsec:0, loc:(*time.Location)(0xc2000f41e0)},
Author:xml.Person{Name:"email-address-removed", URI:"", Email:"",
InnerXML:"<name>email-address-removed</name>"},
Summary:xml.Text{Type:"html", Body:"\n An attempt at adding pubsubhubbub
support to Rietveld.\nhttp://
code.google.com/p/pubsubhubbub\nhttp://code.google.com/p/rietveld/issues/detail?id=155\n\nTheserver
side of the protocol is trivial:\n 1. add a &lt;link
rel=&quot;hub&quot; href=&quot;hub-server&quot;&gt; tag to all\n feeds
that will be pubsubhubbubbed.\n 2. every time one of those feeds changes,
tell the hub\n with a simple POST request.\n\nI have tested this by
adding debug prints to a local hub\nserver and checking that the server got
the right publish\nrequests.\n\nI can&#39;t quite get the server to work,
but I think the bug\nis not in my code. I think that the server expects to
be\nable to grab the feed and see the feed&#39;s actual URL in\nthe link
rel=&quot;self&quot;, but the default value for that drops\nthe :port from
the URL, and I cannot for the life of me\nfigure out how to get the Atom
generator deep inside\ndjango not to do that, or even where it is doing
that,\nor even what code is running to generate the Atom feed.\n(I thought
I knew but I added some assert False statements\nand it kept
running!)\n\nIgnoring that particular problem, I would appreciate\nfeedback
on the right way to get the two values at\nthe top of feeds.py marked
NOTE(rsc).\n\n\n"}}, xml.Entry{Title:"rietveld: correct tab handling\n",
Id:"urn:md5:0a2a4f19bb815101f0ba2904aed7c35a",
Link:[]xml.Link{xml.Link{Rel:"\x00\x18\x0f\x00\xc2\x00\x00\x00\x01",
Href:"\x00f\x0f\x00\xc2\x00\x00\x00\x01\x00\x00\x00\x00\x00\x00\
x00w.appspot.com/124106"}}, Updated:time.Time{sec:63390207737, nsec:0,
loc:(*time.Location)(0xc2000f4240)},
Author:xml.Person{Name:"email-address-removed", URI:"", Email:"",
InnerXML:"<name>email-address-removed</name>"},
Summary:xml.Text{Type:"html", Body:"\n This fixes the buggy tab rendering
that can be seen
at\nhttp://codereview.appspot.com/116075/diff/1/2\n\nThefundamental
problem was that the tab code was\nnot being told what column
the text began in, so it\ndidn&#39;t know where to put the tab stops.
Another problem\nwas that some of the code assumed that string
byte\noffsets were the same as column offsets, which is only\ntrue if there
are no tabs.\n\nIn the process of fixing this, I cleaned up the
arguments\nto Fold and ExpandTabs and renamed them Break and\n_ExpandTabs
so that I could be sure that I found all the\ncall sites. I also wanted to
verify that ExpandTabs was\nnot being used from outside
intra_region_diff.py.\n\n\n"}}}}
want xml.Feed{XMLName:xml.Name{Space:"http://www.w3.org/2005/Atom",
Local:"feed"}, Title:"Code Review - My issues", Id:"
http://codereview.appspot.com/", Link:[]xml.Link{xml.Link{Rel:"alternate",
Href:"http://codereview.appspot.com/"}, xml.Link{Rel:"self", Href:"
http://codereview.appspot.com/rss/mine/rsc"}},
Updated:time.Time{sec:63390216958, nsec:0,
loc:(*time.Location)(0xc2000881e0)}, Author:xml.Person{Name:"rietveld<>",
URI:"", Email:"", InnerXML:"<name>rietveld&lt;&gt;</name>"},
Entry:[]xml.Entry{xml.Entry{Title:"rietveld: an attempt at pubsubhubbub\n",
Id:"urn:md5:134d9179c41f806be79b3a5f7877d19a",
Link:[]xml.Link{xml.Link{Rel:"alternate", Href:"
http://codereview.appspot.com/126085"}}, Updated:time.Time{sec:63390216958,
nsec:0, loc:(*time.Location)(0xc200088240)},
Author:xml.Person{Name:"email-address-removed", URI:"", Email:"",
InnerXML:"<name>email-address-removed</name>"},
Summary:xml.Text{Type:"html", Body:"\n An attempt at adding pubsubhubbub
support to Rietveld.\nhttp://
code.google.com/p/pubsubhubbub\nhttp://code.google.com/p/rietveld/issues/detail?id=155\n\nTheserver
side of the protocol is trivial:\n 1. add a &lt;link
rel=&quot;hub&quot; href=&quot;hub-server&quot;&gt; tag to all\n feeds
that will be pubsubhubbubbed.\n 2. every time one of those feeds changes,
tell the hub\n with a simple POST request.\n\nI have tested this by
adding debug prints to a local hub\nserver and checking that the server got
the right publish\nrequests.\n\nI can&#39;t quite get the server to work,
but I think the bug\nis not in my code. I think that the server expects to
be\nable to grab the feed and see the feed&#39;s actual URL in\nthe link
rel=&quot;self&quot;, but the default value for that drops\nthe :port from
the URL, and I cannot for the life of me\nfigure out how to get the Atom
generator deep inside\ndjango not to do that, or even where it is doing
that,\nor even what code is running to generate the Atom feed.\n(I thought
I knew but I added some assert False statements\nand it kept
running!)\n\nIgnoring that particular problem, I would appreciate\nfeedback
on the right way to get the two values at\nthe top of feeds.py marked
NOTE(rsc).\n\n\n"}}, xml.Entry{Title:"rietveld: correct tab handling\n",
Id:"urn:md5:0a2a4f19bb815101f0ba2904aed7c35a",
Link:[]xml.Link{xml.Link{Rel:"alternate", Href:"
http://codereview.appspot.com/124106"}}, Updated:time.Time{sec:63390207737,
nsec:0, loc:(*time.Location)(0xc2000882a0)},
Author:xml.Person{Name:"email-address-removed", URI:"", Email:"",
InnerXML:"<name>email-address-removed</name>"},
Summary:xml.Text{Type:"html", Body:"\n This fixes the buggy tab rendering
that can be seen
at\nhttp://codereview.appspot.com/116075/diff/1/2\n\nThefundamental
problem was that the tab code was\nnot being told what column
the text began in, so it\ndidn&#39;t know where to put the tab stops.
Another problem\nwas that some of the code assumed that string
byte\noffsets were the same as column offsets, which is only\ntrue if there
are no tabs.\n\nIn the process of fixing this, I cleaned up the
arguments\nto Fold and ExpandTabs and renamed them Break and\n_ExpandTabs
so that I could be sure that I found all the\ncall sites. I also wanted to
verify that ExpandTabs was\nnot being used from outside
intra_region_diff.py.\n\n\n"}}}}
FAIL
FAIL encoding/xml 0.051s

maybe you can upload a newer version of your patch here and I can verify if
I've made any errors in merging.

wrote:
failure
You are right, the test does fail. I am getting the same results when
running from command line like this:

while true; do ./xml.test; done

https://codereview.appspot.com/6114046/

The cause of the problem is in package "reflect":

http://code.google.com/p/go/source/browse/src/pkg/reflect/value.go?r=2c8a88c1efce4e4d78d75b5229d27e4c04031ff7#1494
http://code.google.com/p/go/source/browse/src/pkg/reflect/value.go?r=2c8a88c1efce4e4d78d75b5229d27e4c04031ff7#1902

The garbage collector found the *[]byte and interpreted its contents as
bytes instead of []xml.Link.

Fix: https://codereview.appspot.com/7000059/

https://codereview.appspot.com/6114046/