X-Authentication-Warning:	delorie.com: mail set sender to djgpp-bounces using -f
X-Recipient:	djgpp AT delorie DOT com
RazorGate-KAS:	Status: not_detected
RazorGate-KAS:	Rate: 0
RazorGate-KAS:	Envelope from:
RazorGate-KAS:	Version: 5.5.3
RazorGate-KAS:	LuaCore: 80 2014-11-10_18-01-23
	260f8afb9361da3c7edfd3a8e3a4ca908191ad29
RazorGate-KAS:	Lua profiles 69136 [Nov 12 2014]
RazorGate-KAS:	Method: none
Subject:	Re: DJGPP and Win10 32-bit
To:	djgpp AT delorie DOT com
References:	<56BF484D DOT 4080200 AT iki DOT fi>
From:	"Andris Pavenis (andris DOT pavenis AT iki DOT fi) [via djgpp AT delorie DOT com]" <djgpp AT delorie DOT com>
Message-ID:	<56C0A4F4.4020709@iki.fi>
Date:	Sun, 14 Feb 2016 18:01:56 +0200
User-Agent:	Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101
	Thunderbird/38.5.0
MIME-Version:	1.0
In-Reply-To:	<56BF484D.4080200@iki.fi>
Reply-To:	djgpp AT delorie DOT com
Errors-To:	nobody AT delorie DOT com
X-Mailing-List:	djgpp AT delorie DOT com
X-Unsubscribes-To:	listserv AT delorie DOT com

On 02/13/2016 05:14 PM, Andris Pavenis (andris DOT pavenis AT iki DOT fi) [via djgpp AT delorie DOT com] wrote:
> Installed Win10 Home 32 (201511) bit in VirtualBox VM to try some testing.
>
> The GCC stability problems are the same as in earlier tests on Win10 Enterprise trial (random 
> data corruption in pre-processor, as far as I noticed single byte gets corrupted which often 
> leads to following compiler error, retrying usually succeeds without failure). There is in 
> average 1 compile failure per several hunderds of compiles.
>
> Summary of earlier tests:
> - gcc-3.4.6 and gcc-4.4.7 are OK (earlier installed gcc-3.4.6 rebuilt it on W10, bootstrapped 
> gcc-4.4.7 with 3.4.6 and all worked without problems)
> - trying to bootstrap 4.6.4 with gcc-4.4.7 run into similar trouble in bootstrap stage 3 (there 
> were no problems in stage 1 and 2)
> - all later versions 4.7.4 and newer have the same problem
> - failed to reproduce similar problems with different programs
>
> Some thoughts:
> - increase of size of executable (cc1.exe about 8.5MB for 4.4.7 and 21MB for 5.3.0)
It could be the reason.
> - I got corruption in pre-procesed output. So that should be easy in comparison with compiling 
> itself as compiling require noticeably more resources. That points that the actual problem could 
> also be on our side
> - It came into mind that the problems began in the same versions in which we had to handle 
> relocation counter overflow in 2011. I used then some tricks of changing compiler options to 
> avoid overflow then with gcc-4.6 (support in binutils were implemented after that). I guess I 
> could try to build gcc-4.6.4 both with changed compiler options and without changes and see 
> whether there is any change and perhaps also test with old binutils version (2.21).
Got it also with binutils-2.21.1 and STAGE1_CFLAGS=-Os BOOT_CFLAGS=-Os CXXFLAGS=-Os when building 
gcc-4.6.4. although not very often (2 times per bootstrap). gcc-6.0.0-20160203 is much worse.
Did some additional stress testing of memory allocation/deallocation in random order:
- this time filling allocated random size blocks with random data
- calculating and checking CRC32 of both block contents and information about block
Allocated total size is about 800-900MB, simultaneously block count perhaps 15000000 - 25000000.
Stress test now running on WIn10 Home 32 bit for more than a hour and no errors detected.
>
> New unrelated problem:
> - this time I created 2 partitions in VM
> - all works as earlier on C:
> - there are LFN related problems with drive D: - if directory names are SFN only, all is OK, 
> otherwise attempt to open file in such directory fails with EACCES ('cat d:/foo/example.txt' is 
> OK, 'cat d:/foo.example/bar/example.txt' fails with EACCES, assuming that files exist). I did not 
> have drive D: in earlier tests so I do not know whether the problem is Win10 Home specific.
>
Seems to be related to use of NTFS. Formated partition as FAT32 and after that no more problems on 
D:. One only needs to use external tools to format FAT32 as Windows 10 Home do not seem to offer 
such possibility.

Andris

- Raw text -