yak1ex / ax7z Goto Github PK

Current behavior of loading 7z.dll:

• Filename only (default DLL search path)
• Path from HKCR
• Path from HKLM

It is necessary to change the order and to eliminate current directory.

• Consider folder structure
• At first, minimum required only, then switching
• Consider solid block

vector<char> is not correctly terminated with '\0'. It happens to work without application errors. Maybe pop_back() loop is to get directory part of the current module but this is contradictory to the code afterwards to append just ".ini". Anyway, the current ini file path, which is a spi path + ".ini", should be preserved.

reported by http://8904.teacup.com/yak_ex/bbs/135

There are 2 problems.

• Boost.Thread uses GetTickCount64() if _WIN32_WINNT >= 0x0600
• VS 2012 SP3 is necessary to build binary working on XP
  http://tedwvc.wordpress.com/2012/03/11/how-to-get-visual-c-2012-vc-11-beta-statically-linked-crt-and-mfc-applications-to-run-on-windows-xp/
  http://tedwvc.wordpress.com/2013/04/14/how-to-get-visual-c-2012-update-2-statically-linked-applications-to-run-on-windows-xp/

Here are the corresponding solutions.

• Specify _WIN32_WINNT is 0x0501 (XP) explicitly.
• Update VS 2012

Boost.Thread has some fixes for Windows environment. So, it might get better behavior.

Should be fixed for both branches of plain and master.

1.54 has BOOST_THREAD_VERSION = 4. Right now default is 2 but it is better to immigrate on this occasion.

Reproduced behavior like mentioned in http://gorry.haun.org/diary/direct.cgi?2008100802

Cherry pick to common then rebase?

Only the first image is extracted and progress dialog is shown repeatedly.

Store an ini file to AppData as default for per-user configuration.
Users can select to store an ini file to the folder to contain DLL as before.

• not-yet-declared template type parameter
• In C++98, map::erase() doesn't return iterator
• boost::thread::id via ...

try_join_for() in Queue::CleanupAll() has the following behavior on my laptop.

• try_join_for(boost::chrono::milliseconds(0) -> works but seems not to wait
• try_join_for(boost::chrono::milliseconds(1) -> blocked without 1st image
• try_join_for(boost::chrono::milliseconds(2) -> blocked completely
• try_join_for(boost::chrono::milliseconds(3) -> works
• try_join_for(boost::chrono::milliseconds(5) -> works
• try_join_for(boost::chrono::milliseconds(100) -> works
• try_join_for(boost::chrono::milliseconds(500) -> works but too slow

Maybe instability of steady_clock() on Win32 affects it. (see https://svn.boost.org/trac/boost/ticket/8323 )

Instead try_join_until(system_clock::now() + d) might be stabler.

Currently problems exceptions are thrown are totally ignored. Log and show a message to a user for exceptional cases.

SolidCacheDisk::SetCacheFolder() uses sqlite_open16() so that unintentional encoding conversion occurs query/insert for text. It causes cache records for archives whose paths contain non-ascii letters are available only until the next time of DLL loading. This leads advantages of persistent cache to very limited effects.

On the other hand, SolidCacheDisk()::Clear() uses sqlite_open() so that persistent cache records are kept.

It is necessary not only to unify open scheme but also to check "pragma encoding" in SolidCacheDisk()::InitDB().