Case-sensitivity in a filesystem is about the dumbest idea in Unix
When file names are considered just as sequences of bytes with no special meaning (except for 0 and 47 '/'), all one has to do in order to tell if two names are equal is a very simple (and fast) byte comparison, and you don't have to deal with encodings, equivalence classes of characters and conversions, or the way to represent '..' of the day.
Ever wondered why URIs are case-sensitive, or why case-insensitive strings on the net are tipically limited to ASCII characters (or a subset of ASCII)? Being case insensitive requires a lot of additional work, and buys you nothing since ordinary Joe can select his files in a nice file dialog with no typing at all.
Next, allow deletion of files that are in use? Why?
This way you can have:
absolutely private temporary files/directories that cannot be tampered with, and which are automatically erased when the application terminates/dies
smooth upgrades of programs and libraries without disturbing running applications
(deleted data is still accessible to processes currently using it). This, for example, permits remote upgrades to ssh even if the only way to reach the host is ssh itself, without having to close the connection (so if one screws something up, he has still a chance to fix things before having to take the car and get to the server room).
FireFox, by default, requires you to whitelist sites to install software from them. So, no exploits from that side
But being installed from a trusted source does not tell the whole story about being able to completely trust an extension: it could have bugs resulting in exploitable security flaws (that has been the case with GreaseMonkey - now fixed), and NoScript doesn't block trusted code (it can't, otherwise the GUI wouldn't work anymore). Should it ever happen with a popular extension (like AdBlock or Googlebar), things could get pretty bad.
Please note that this very same logic applies to IE as well, and more in general to every piece of software with nontrivial third-party plugins.
Actually, there is some effort going on to ensure that Mozilla extensions (and trusted code in general) can handle unsafe content only in a properly sandboxed enviroment, so I wouldn't dismiss the argument as a closed case yet.
It would be nice if a full Gnome X-session would work across my LAN. Slow as molasses
Here it always worked as a charm (and there are more than 300 hosts on this LAN, mostly Windows XP machines used as fancy X terminals -- don't ask why -- and Linux and HP-UX boxes), in fact I use it as my primary desktop (with fullscreen XWin.exe).
If it goes "slow as molasses", your LAN has definitively problems (remember, X protocol is more sensitive to latency problems than bandwidth problems), or it is not really a LAN (and in that case you should really use VNC, or Nomachine NX).
That was true ages ago, when it was either SVGAlib or nothing (and since SVGAlib was a library that needed direct access to hardware, binaries linked against it needed to be suid root in order to work).
Since then, basically every game out there that used SVGAlib has been ported to either libSDL or libclan (Abuse included), which use accelerated X as the default backend, thus requiring no root privileges.
As long as it's without the adds, I think it's great to see what i still consider the best browser to be present on cell phones. Albeit that probably means the cell phones are running CE...
Well, high-end Nokia phones run Symbian OS (obviously: Nokia is a notable member of the Symbian consortium...), which basically is the good 'ol Epoc 32 which ran on Arm PDAs like the Psion Series 5. And Opera has been running for years on such machines, so that should be a no-brainer.
OTOH, Nokia uses its own GUI on top of Symbian, so this will probably mean some minor adjustments for Opera.
GNU grep has no arbitrary limit on line size, except the available memory. Most GNU utilities do not impose arbitrary limits like their *NIX counterparts, and they are only limited by the available memory which can be allocated (which you can furtherly limit via ulimit, btw).
grep matches regular expressions against one line at a time
to do that, it has to read a whole line first
It is obvious that the line (which is of infinite length since it comes from/dev/zero) does not fit in the available memory, so GNU grep rightly fails.
I fail to see why one can't have a zero button mouse that simply executes the appropriate action after a predefined delay.
Other replies aside ("midas touch", etc.), I'd like this for switching between tabs, being it the logical extension of "focus follows pointer".
Firefox has an extension which does this ("Tabbrowser preferences"), and I found myself trying to do the same thing on GTK and Qt apps.
It could be an interesting addition to these toolkits (configurable, of course, perhaps disabled by default, like it already is for detachable menu bars and such).
Re:Linus is probably biased about Mach though....
on
Torvalds Switches to a Mac
·
· Score: 2, Insightful
in addition i think linux is not completely monolithic anymore and has become a lot more modular to the Mr. Tanenbaum's liking
Well said, but the real point about microkernels is to have distinct kernel subsystems in distinct address spaces, so a pointer gone wild in one subsystem can't corrupt data in other subsystems.
OTOH, a single address space means that coders have to triple check their code and pay attention to side effects, which is not a bad idea after all (provided you are not pressed by release dates, as it's the case for the Linux kernel).
Can someone please give examples of when a book converted to a movie was anywhere near as good as the book? Some are satisfactory conversions... but I have never had one instance where a good book became a better movie...
Blade Runner (althought it can be said that it hasn't that much to share with the novel by Philip Dick). Clockwork Orange (ok, Kubrick had the USA edition without the ending chapter). The Andromeda Strain (the movie and the book are pretty much the same).
Making a good transposition from novel to movie is hard because:
The length of a movie is just that, in the range of a couple of hours. A novel can go deep in all sort of non-visual details, some of which cannot be transposed as flashbacks: you have to leave them out in the movie
action in movies needs to be spread evenly (you can't have all action at the end or at the start, for example), otherwise you get a boring movie. Novels can go for hundred of pages without any (visually representable) action, and still be interesting.
the inner flows of thoughts of characters are hard to be transposed in a movie: either they become dialogues (that weren't dialogues in the novel), or you need some out-of-frame narrating voice (which quickly gets boring in a movie)
Now... we''ve also got the alarmingly bad Select() method
Which is strange, because (by your description) it has exactly the same shortcomings of the old *NIX select(2) system call (that's why poll(2) is there). One would expect that people designing a library in the 21st century knew better than this.
With my heart condition at birth I myself would not be alive if it wasn't for modern medicine and surgery. I find my self wondering if it's fair for me to have a child with my genes...
In a technological society, traits transmitted mainly by education (intelligence, behaviour) count by far more than physical traits transmitted mainly by genes, because the formers can obviate to deficiencies of the latters in ways that are infinitely faster than the ones which apply in the opposite case (at least that's true for humans). Nature may be wise in preserving life in general, but humans can be wiser in preserving human life.
And what "power" does ANSI or ISO have compared to the companies that implement their standards?
Because pratically all governments require compliance to said standards before dealing with you, and contracts with governments usually involve large sums of money that you can count on being paid to you sooner or later.
Sorry, but HTML and CSS are not standards in any way: they are recommendations from a consortium that on the whole has very little power compared to the one held by other players in the field.
(X)HTML and CSS are not de iure standards, since any half-baked rendering engine out there may claim to be supporting both even if it is not completely true, without fearing the visit of the W3C cops telling "sorry, you can't publicly state that unless your product passes ALL our validations tests".
And as de facto standards, there are still strong issues: just look at the "quirks" mode in Gecko (which emulates IE behaviour for sloppy-coded pages).
That being said, a concise recommendation is light years better than an implementation without sources (which has to be reverse-engineered to be reimplemented), so if you want to blame Microsoft, blame it for not providing a concise description of the IE behaviour (the reason behind this being that an IE-only web meant a Windows-only web).
Yet whenever someone complains about an option being removed from the main config dialogs, the standard response is, "use GConf." So what is it? Are we supposed to use GConf or not?
It's the same with about:config in Firefox: if you need it to set some option, you know it's there, but there are also extensions that will do that for you with a prettier GUI.
For Gnome it's just a matter of installing something like GTweakUI instead of resorting to gconf-editor.
A curiosity: some pages (see page 781 and subsequent) are stil used today to broadcast data (mostly old MS-DOS binaries, and textfiles from the italian branch of project Gutemberg) to PC equipped with a special decoding card.
That said, I think there's a lot to be considered when choosing a software project name. One that quickly and accurately describes or at the very least gives a hint as to the functionality of the package is great.
That's at the filesystem level. I think they mean at the device driver level.
Well, since access at the device driver level from userland is still done via special char/block device files, there isn't really that much difference.
Just ensure that all user-mountable entries (if any) in/etc/fstab have options "nosuid" and "nodev", and your char/block devices in have proper permissions.
Give me nested directories 30 levels deep!! And no spatial browsing please!
I agree with you that putting everything in the same basket and rely just on metadata to extract info is quite stupid (probably we'll end with a "folder" and "subfolder" metadata anyways, so what's the point?). But relying exclusively on directory trees to classify data has definitively limits, in that a certain file may belong to at most one category (if you are thinking that symlinks and hardlinks solve that problem nicely, please consider that symlinks break if you rename/move the original file, and hard links are limited to the same volume, and there is high inconvenience in making a proper backup of them - in that you end with multiple copies of the same data).
That being said, one could say there's no real need for special filesystem support to store/search metadata: let's do a periodic collection of metadata from files, just like locatedb does, and we are all happy, right? Well, for one there is the latency between updates, and there are also are issues in a multi-user environment (one should see results just for files which he may access at the instant of the query).
The right answer is probably somewhere in the middle, I'd say that ReiserFS 4 with its plugin architecture is a great tool to experiment with this (idea: each time a file that was opened for writing/appending is closed, notify an userspace daemon to extract its metadata and put a copy of it somewhere: metadata is still in the file, and its copy is kept up-to-date in a form which is easier to perform searches on using an uniform API, but you need some kernel support to do this efficiently).
Re:People still use a shell for Linux?
on
Bash 3.0 Released
·
· Score: 2
Bad example. Using KDE, I click on my home directory icon, select the images I want to convert, right-click on one of them and pick Actions | Convert To | PNG.
Ok, now do it every morning, at 3am, except on Sundays, and move corrupted images to the trashcan.
The point of a shell script is not to automate one-time generic tasks (that's what your context menu entry is for), but to automate one-time higly specific tasks (for which you can't reasonabily expect to have a predefined context menu entry) or unattended tasks that have to be executed many-times.
In the end, scripting everything is dumb, but requiring your presence to do the same task by hand over and over is dumber. You know the saying, "be quiet or I'll replace you with a very small shell script", there is a reason behind that.
Slashdot Mirror
When file names are considered just as sequences of bytes with no special meaning (except for 0 and 47 '/'), all one has to do in order to tell if two names are equal is a very simple (and fast) byte comparison, and you don't have to deal with encodings, equivalence classes of characters and conversions, or the way to represent '..' of the day.
Ever wondered why URIs are case-sensitive, or why case-insensitive strings on the net are tipically limited to ASCII characters (or a subset of ASCII)? Being case insensitive requires a lot of additional work, and buys you nothing since ordinary Joe can select his files in a nice file dialog with no typing at all.
Next, allow deletion of files that are in use? Why?
This way you can have:
When properly indented, it seems quite reasonable. What really scaries me is the <w:validateAgainstSchema/> <w:saveInvalidXML w:val="off"/> bits...
But being installed from a trusted source does not tell the whole story about being able to completely trust an extension: it could have bugs resulting in exploitable security flaws (that has been the case with GreaseMonkey - now fixed), and NoScript doesn't block trusted code (it can't, otherwise the GUI wouldn't work anymore). Should it ever happen with a popular extension (like AdBlock or Googlebar), things could get pretty bad.
Please note that this very same logic applies to IE as well, and more in general to every piece of software with nontrivial third-party plugins.
Actually, there is some effort going on to ensure that Mozilla extensions (and trusted code in general) can handle unsafe content only in a properly sandboxed enviroment, so I wouldn't dismiss the argument as a closed case yet.
Here it always worked as a charm (and there are more than 300 hosts on this LAN, mostly Windows XP machines used as fancy X terminals -- don't ask why -- and Linux and HP-UX boxes), in fact I use it as my primary desktop (with fullscreen XWin.exe).
If it goes "slow as molasses", your LAN has definitively problems (remember, X protocol is more sensitive to latency problems than bandwidth problems), or it is not really a LAN (and in that case you should really use VNC, or Nomachine NX).
The advisory on SecurityFocus.
That was true ages ago, when it was either SVGAlib or nothing (and since SVGAlib was a library that needed direct access to hardware, binaries linked against it needed to be suid root in order to work).
Since then, basically every game out there that used SVGAlib has been ported to either libSDL or libclan (Abuse included), which use accelerated X as the default backend, thus requiring no root privileges.
Well, high-end Nokia phones run Symbian OS (obviously: Nokia is a notable member of the Symbian consortium...), which basically is the good 'ol Epoc 32 which ran on Arm PDAs like the Psion Series 5. And Opera has been running for years on such machines, so that should be a no-brainer.
OTOH, Nokia uses its own GUI on top of Symbian, so this will probably mean some minor adjustments for Opera.
Other replies aside ("midas touch", etc.), I'd like this for switching between tabs, being it the logical extension of "focus follows pointer".
Firefox has an extension which does this ("Tabbrowser preferences"), and I found myself trying to do the same thing on GTK and Qt apps.
It could be an interesting addition to these toolkits (configurable, of course, perhaps disabled by default, like it already is for detachable menu bars and such).
in addition i think linux is not completely monolithic anymore and has become a lot more modular to the Mr. Tanenbaum's liking
Well said, but the real point about microkernels is to have distinct kernel subsystems in distinct address spaces, so a pointer gone wild in one subsystem can't corrupt data in other subsystems.
OTOH, a single address space means that coders have to triple check their code and pay attention to side effects, which is not a bad idea after all (provided you are not pressed by release dates, as it's the case for the Linux kernel).
Blade Runner (althought it can be said that it hasn't that much to share with the novel by Philip Dick). Clockwork Orange (ok, Kubrick had the USA edition without the ending chapter). The Andromeda Strain (the movie and the book are pretty much the same).
Making a good transposition from novel to movie is hard because:
They probably need to ask the Mozilla Foundation because of the trademarks.
Which is strange, because (by your description) it has exactly the same shortcomings of the old *NIX select(2) system call (that's why poll(2) is there). One would expect that people designing a library in the 21st century knew better than this.
In other words: if you aren't dedicated to it full-time, you can't make art? Hmmmm...
To me it looks INTENTIONALLY cheap, to make sure that even oog the caveman can understand it is a prerelease not meant for general use.
In a technological society, traits transmitted mainly by education (intelligence, behaviour) count by far more than physical traits transmitted mainly by genes, because the formers can obviate to deficiencies of the latters in ways that are infinitely faster than the ones which apply in the opposite case (at least that's true for humans). Nature may be wise in preserving life in general, but humans can be wiser in preserving human life.
Because pratically all governments require compliance to said standards before dealing with you, and contracts with governments usually involve large sums of money that you can count on being paid to you sooner or later.
(X)HTML and CSS are not de iure standards, since any half-baked rendering engine out there may claim to be supporting both even if it is not completely true, without fearing the visit of the W3C cops telling "sorry, you can't publicly state that unless your product passes ALL our validations tests".
And as de facto standards, there are still strong issues: just look at the "quirks" mode in Gecko (which emulates IE behaviour for sloppy-coded pages).
That being said, a concise recommendation is light years better than an implementation without sources (which has to be reverse-engineered to be reimplemented), so if you want to blame Microsoft, blame it for not providing a concise description of the IE behaviour (the reason behind this being that an IE-only web meant a Windows-only web).
It's the same with about:config in Firefox: if you need it to set some option, you know it's there, but there are also extensions that will do that for you with a prettier GUI.
For Gnome it's just a matter of installing something like GTweakUI instead of resorting to gconf-editor.
For running a typesetting system for patent applications?
No, really!.
A curiosity: some pages (see page 781 and subsequent) are stil used today to broadcast data (mostly old MS-DOS binaries, and textfiles from the italian branch of project Gutemberg) to PC equipped with a special decoding card.
For example: "This Program Lets You Write Letters".
:-)
Well, since access at the device driver level from userland is still done via special char/block device files, there isn't really that much difference. Just ensure that all user-mountable entries (if any) in /etc/fstab have options "nosuid" and "nodev", and your char/block devices in have proper permissions.
I agree with you that putting everything in the same basket and rely just on metadata to extract info is quite stupid (probably we'll end with a "folder" and "subfolder" metadata anyways, so what's the point?). But relying exclusively on directory trees to classify data has definitively limits, in that a certain file may belong to at most one category (if you are thinking that symlinks and hardlinks solve that problem nicely, please consider that symlinks break if you rename/move the original file, and hard links are limited to the same volume, and there is high inconvenience in making a proper backup of them - in that you end with multiple copies of the same data).
That being said, one could say there's no real need for special filesystem support to store/search metadata: let's do a periodic collection of metadata from files, just like locatedb does, and we are all happy, right? Well, for one there is the latency between updates, and there are also are issues in a multi-user environment (one should see results just for files which he may access at the instant of the query).
The right answer is probably somewhere in the middle, I'd say that ReiserFS 4 with its plugin architecture is a great tool to experiment with this (idea: each time a file that was opened for writing/appending is closed, notify an userspace daemon to extract its metadata and put a copy of it somewhere: metadata is still in the file, and its copy is kept up-to-date in a form which is easier to perform searches on using an uniform API, but you need some kernel support to do this efficiently).
Ok, now do it every morning, at 3am, except on Sundays, and move corrupted images to the trashcan.
The point of a shell script is not to automate one-time generic tasks (that's what your context menu entry is for), but to automate one-time higly specific tasks (for which you can't reasonabily expect to have a predefined context menu entry) or unattended tasks that have to be executed many-times.
In the end, scripting everything is dumb, but requiring your presence to do the same task by hand over and over is dumber. You know the saying, "be quiet or I'll replace you with a very small shell script", there is a reason behind that.