Re: New installer fails miserably when path of %TEMP% contains spaces
Jernej Simončič
> Really?
> 7-zip CANT be used on a Windows before you install it. And before that, you
> have to download it. Which might be a little tricky without internet access.
> MAKECAB.EXE etc. are all on board and available without installation.
Who cares? Only the developers need 7-zip installed on their machine, the
end user gets a self-extracting file which he just needs to run.
> And: what's the overall disk space consumption of the new opera installer?
> How many MBs and files/folders get shuffled around?
> - about 7MB for the self-extracting archive (1 file);
> - about 18MB in %TEMP% (13 folders, 137 files);
> - about 18MB in the destination directory (13 folders, 137 files).
Disk space on the end user's machine doesn't really matter when you can get
a 2TB drive for less than 80€. What matters is the amount needed to
download since bandwidth isn't free for the uploader.
> Compare these numbers against a native Windows installer built with *.INF
> and *.CAB:
> - about 7MB for the *.CAB and *.INF (2 files);
> - about 18MB in the destination directory (13 folders, 137 files).
Bzzt - you have 2 files - how will you ship them to the user? Remember,
people are used to download a single file to install something. Of course,
you could pack those 2 files to another archive, but that kinda defeats the
purpose. Also, with CAB+INF you can't just double-click one file to install
- you have to know that you need to right-click the inf file and select
Install - but the majority of users will never bother with that.
> About 18MB less disk space (NO temporary space needed at all for intermediate
> decompression/extraction of the whole archive!), about 36MB less IO (the meta
> data for creation/removal of the temporary files/folders not counted), and a
> shorter installation time too!
> Can 7-zip (or any other tool) beat that?
Yes, just about any installer can (besides, you're looking at the wrong
things - what matters for the publisher is to minimize bandwidth
consumption, and what matters to the user is to have as simple install
process as possible) - except that none of them are flexible enough for
Opera to use.
>> (even when you count the overhead of decompression code)?
> Do you count the overhead of downloading and installation of 7-zip by the end
> user?
No, because they don't need it.
>> There's a very good reason why a ton of programs pack their installers with
>> 7-zip for redistribution.
> No. There is never good reason not to use the platforms native tools.
When's the last time you saw an installer that uses inf+cab method?
Follow-up to opera.off-topic.
--
begin .sig
< Jernej Simončič ><>◊<>< jernej|s-ng at eternallybored.org >
end
Jernej Simončič
>> What matters is the amount needed to
>> download since bandwidth isn't free for the uploader.
> If true, then *.CAB without a stub for extraction is the smallest form for
> distribution.
I extracted Opera_installer_1100_1060_en.exe (7110051 bytes) to a new
directory, then compressed it to a CAB file (which turned out to be a much
more complex task than creating a typical archive). Anyway, the result of
running
MAKECAB.EXE /D CompressionType=LZX /D CompressionLevel=7 /D CompressionMemory=21 /f opera.txt
is a 7480130 bytes large CAB file - 370079 bytes larger than the original
installer (and this is despite the original installer having a nearly half
megabyte stub thanks to a huge uncompressed icon it includes).
>> Bzzt - you have 2 files - how will you ship them to the user? Remember,
>> people are used to download a single file to install something.
> Unfortunately, people have been dumbed down.
No, the computers were simply made more accessible to normal people (which
just want them to work, without having to read through several 500-page
manuals first).
> If you REALLY want to have an *.EXE use IEXPRESS.EXE and create one.
> This will extract *.CAB and *.INF to %TEMP%, then run the *.INF, and still
> use less resources than the 7-zip SFX.
Except that a plain CAB file is already larger than the current installer.
>>> About 18MB less disk space (NO temporary space needed at all for intermediate
>>> decompression/extraction of the whole archive!), about 36MB less IO (the meta
>>> data for creation/removal of the temporary files/folders not counted), and a
>>> shorter installation time too!
>>> Can 7-zip (or any other tool) beat that?
>> Yes, just about any installer can
> Show the numbers!
Basic Inno Setup installer 7429332 bytes - still smaller than the CAB file,
and much more user-friendly at the same time. I won't bother with NSIS,
because it's scripting is much more complex, which is too time consuming
just to prove a point (but NSIS usually creates smaller files than Inno).
>> (besides, you're looking at the wrong things - what matters for the publisher
>> is to minimize bandwidth consumption,
> Where 7-zip has NO advantage over MakeCAB!
Already proven to be false.
>> When's the last time you saw an installer that uses inf+cab method?
> At least every time I install a device driver, a Windows component or a
> Windows update!
Drivers either use their own installer, or come packed in ZIP files (and
are then used by the Add new hardware wizard). Their INF files don't
support the right-click -> Install method of installation. Windows updates
use .msu files, which is a custom format (updates for older versions of
windows come in self-extracting files - again no CAB and INF files in
sight).
> Smart developers use *.CAB and *.INF^W^W^Wtheir platforms native format.
> Others just haven't learned...
I guess there's no smart developers on Windows then?
Jernej Simončič
>> I extracted Opera_installer_1100_1060_en.exe (7110051 bytes) to a new
>> directory, then compressed it to a CAB file (which turned out to be a much
>> more complex task than creating a typical archive).
> You might have used CABARC.EXE which doesnt need a .DDF (diamond directive
> file) as input.
I could? But that needs the platform SDK or MS cabinet tools, neither of
which are installed on every machine by default (plus cabarc creates larger
files for some reason, despite using -m LZX:21).
> .CAB dont need an extractor since it is the native format, 7-zip bit but does,
> so you'll have to account for the stub.
I did account for the stub - and the file was still 361kB smaller.
> Opera_installer_1100_1094_en.exe : 7217606 bytes
> ------- = 1 : 1,0369 ~ 27 / 28
> opera11.cab : 7483986 bytes
> The difference is barely noticable, just within the typical error margin.-)
Relative difference doesn't mean much - 361kB is noticable when you have a
few millions of downloads (especially when the majority of these downloads
try to happen in the first few hours after a release).
>> No, the computers were simply made more accessible to normal people (which
>> just want them to work, without having to read through several 500-page
>> manuals first).
> You need 500 pages for a simple sentence like
>| Download both
99% people won't read past this word (why should they - every other piece
of software only needs them to download a single file and double-click it).
> Never heard of unattended setup? There is no .EXE to run then, and no
> "Add new hardware wizard". Just a list of directories to be given in
> [HKLM\Software\Microsoft\Windows\CurrentVersion] DevicePath=
> and the drivers unpacked into these directories.
That's what the installers do (unless by unattended setup you meant the
drivers that are installed from the OS install media).
>> Their INF files don't support the right-click -> Install method of
>> installation.
> Wrong. You can write INF files for driver installation for both right
> click and the "Add new hardware wizard".
I tried several drivers on my disk, and all of them reported "The INF file
you selected does not support this method of installation". I wonder why.
>> (updates for older versions of windows come in self-extracting files
>> - again no CAB and INF files in sight).
> Wrong once^Wtwice more:
> 1. the .EXE you reference are .CAB and use an UPDATE.INF, interpreted but
> through UPDATE.EXE (for Windows 9x, the .EXE were IEXPRESS installers,
> with .INF for SetupX.Dll, SetupAPI.Dll or AdvPack.Dll).
We must use a different OS - the XP updates I've seen unpack themselves to
directories with long names on a random drive, then run update.exe from
there - no CABs in sight in those directories.
> Wrong guess! Check the drivers for Matrox Graphic cards, for example.
> Or nVidia drivers: their NView is installed from NVIEW.CAB.
...but where's the INF file for NView? Besides, before you get to the NView
installer, the drivers have to be unpacked - from a self-extracting 7-Zip
archive (now where have we seen that before?).
(btw, it was similar with Matrox when I still had my P650 card - their
drivers were distributed in a self-extracting ZIP file)
> Software which I compile or distribute comes in .CAB and uses .INF
Is your software intended for the general public, and if it is, how many
users (or downloads) do you have?