Interesting to see the UI change (degradation) over time. The first version makes it very clear what's a button and clickable at first glance, while the latest version looks like a sea of whitespace and makes that far more difficult.
I started with Visual C++ 1.0. There were no templates.
Visual C++ 2.0 had templates.
There was by Visual C++ 3.0. They went to 4.0 to sync the MFC (Microsoft Foundation Classes) versions.
IIRC, you could use a bit of STL with 4.0
Visual C++ 5.0 was mainly optimizer differences.
Visual C++ 6.0 was actually pretty good. However it lacked partial template specialization. I was a Boost author at the time and lots of Boost code had specific workarounds for Visual C++ 6.
Visual C++ 2002 also had no partial template specialization.
Visual C++ 2003 was the first version with partial template specialization and that could compile all of Boost.
Visual C++ 2005 and 2008 did not have much changes.
Visual C++ 2010 tried to get back to 6 in the IDE (there was a deliberate marketing as such). It also had some C++11 features. But no variadic templates.
Visual C++ 2012 had no variadic templates.
Visual C++ 2013 was the first with variadic templates.
Nowadays, Visual C++’is doing much better tracking the C++ standard and often has compiler and library features before clang and GCC.
It doesn't point out the massive productivity boost of Visual C++ 6, IntelliSense (ie, autocomplete). This saved so much time from having to look up documentation.
I worked on that! I was a dev on the team that built the first IntelliSense engine for C++. It’s a miracle it worked at all. It was based on a hacked-up version of the C++ front-end, but when it inevitably hit errors attempting a single-pass parse of the current source, it would silently bull its way through and try not to get too screwed up. Doing that in the presence of templates was not a good time. But with the RAM and processing speeds available then, I’m still kind of shocked it worked at all, 28 years ago.
heh, sorry, but intellisense was the first thing we (in a game studio) turned off immeadiately after install (by deleting the dll as there was no other way) because it was unacceptable resource hog. this was around VS 2003/2005/2008.
(Not so) funny that there was nothing "visual" in Visual C++. The dialog resource editor was a joke. I remember how I was relieved when I discovered Borland C++Builder, so I didn't have to switch to Delphi for Windows software development.
Even worse, to this day, Microsoft has completely failed to produce anything comparable to C++ Builder due to the internal feuds around C++ and COM.
C++/CLI could have been it (after the Managed C++ reboot), but its integration with Windows Forms has always been clunky, nevermind WPF.
Then C++/CX could have been it, but then there was this rebellion replacing it with C++/WinRT, nowadays in maintenance, only getting bug fixes, being good enough for WinUI 2.0 and WinAppSDK low level stuff and that is about it.
I was big into Borland from their early MS-DOS compilers, all the way into 1999.
Embarcadero might not be Borland, but still manages to put out great GUI RAD tooling, whereas Microsoft lost sight of what means a great developer experience for native applications.
One thing not mentioned is that Visual C++ and Visual Basic historically were separate IDEs with separate codebases. When the time came to unify them, only one of them could continue on. My understanding is that Visual Basic won, and that today's Visual Studio IDE (devenv.exe, msenv.dll, etc.) is the continuation of that VB codebase.
I don't actually know in which release that transition happened. But since there's a screenshot of each version in the article, presumably that transition is visually documented...
That might have been the transition from Visual C++/Basic 4 to 6 (I skipped 5), but the cataclysmic one was the switch from Visual Studio 6 to Visual Studio .NET, when large portions of the IDE and build system were rewritten in .NET. Visual Studio .NET (2002) was much slower and much buggier than VS6. The native debugger was glacial at conditional breakpoints and debug output, the build system took an eternity to do a dependency check, its UI visibly redrew more slowly, etc. It was so bad that Microsoft had to create a special offer for the Visual Studio .NET (2003) upgrade for only ~$30.
This transition was not great for Visual Basic developers either since their language was transitioned from generating native code (VB6) to becoming dependent upon the .NET Framework (VB.NET), supported secondarily to C#.
To this day I would consider VB 6 the epitome of COM development tooling, neither .NET, nor the UWP trainwreck were as good, and C++ tooling only MFC is worth of mention, as something putting development experience first, instead of Windows division politics.
Quite awkward for what is such a key technology on Windows.
If memory serves, it was Visual InterDev that won, actually. The VID IDE was the forerunner to Visual Studio.NET 2002, which was the first unified Visual Studio IDE.
n.b. I worked on the Visual Studio Core team, which maintained devenv.exe, among other things, from 2003-2007'ish.
Visual InterDev was the first IDE I ever used outside of notepad.
It was bundled in with one Microsofts early webediting tools with office (97/200)....but not FrontPage.
I thought I was touching the future, it was so amazing! I really don't remember much about other than the impression that it left on me.
After Visual C++ 6. They broke a lot of the C++ IDE features and they weren’t as good as the prior versions (dialog editor etc) so for a long time we preferred staying on 6. I think if we could have the newer compilers but the snappiness of that UI many developers would be happy.
It’s also a product of the segmentation of the developer tools in Microsoft. The Windows team was responsible for the compiler rather than the Developer Tools team.
No, that's not really true. I was on the C++ compiler team from 1991 to 2006. When I first started, the DevTools team reported up through the Windows team, but never really felt a well-integrated part. We were never in the same building as the Windows team, for instance. I remember, probably 1992 or 1993, driving from building 4 where the compiler team lived to the Windows building (forget which one that was, maybe in the 12 to 15 block back then?) to get a copy of the Windows NT source on a hard drive. That's because I was a dev on the C++ compiler back-end team then (moved to the front-end in '95, IIRC), and compiling that source was a major test of the 32-bit compiler I was working on.
Don't remember when DevTools was re-orged out from under Windows, but I'm pretty sure it was by '95, and well before VC++ 6.
No, there were two devs working on the 68k Mac compiler, with ~10 devs on the x86 side (though both targets shared a lot of code and differed mainly in the late codegen and peephole optimization phases). I never worked on the 16-bit code; the 32-bit and later 64-bit x86 backend was a different codebase from the 16-bit stuff.
If I remember from poking around at the 16 bit version in that timeframe (1.52?) there was still an option to generate P-code and (this is now nearly thirty years ago) that was used by Excel to minimize code size for the segmented 16 bit systems.
Any idea how many devs are working at Microsoft on the C and C++ compilers these days? I've heard rumours that there's more on the Rust team and that C++ is taking a back seat.
No clue. I left as a full-time employee in 2007, did a few contractor gigs with various old teams of mine to help out, but that was done by 2013. I lost touch with how things were going internally after that.
The dialog editor is a good example of the damage that was done with the VS.NET transition. When they rewrote it to use the WinForms-based UI, they introduced a fundamental bug: the Z-order for picking was reversed so that clicking on a stack of controls selected the one on the bottom. Very annoying for controls like group controls intended to be stacked below other controls. Bug filed, WONTFIX'd, and it's still broken to this day.
No, overlapping sibling controls is absolutely supported in Win32. It has a defined Z-order for child controls along with functions to manage it:
> The system maintains the z-order in a single list. It adds windows to the z-order based on whether they are topmost windows, top-level windows, or child windows.
The WS_CLIPSIBLINGS window style is also explicitly for controlling cross-clipping of child windows during painting:
> Clips child windows relative to each other; that is, when a particular child window receives a WM_PAINT message, the WS_CLIPSIBLINGS style clips all other overlapping child windows out of the region of the child window to be updated.
Furthermore, group boxes and tab controls require overlapping child windows, because they are designed for the controls that visually appear within them to be sibling controls that are raised in Z-order. This is why the dialog editor's current behavior is broken, because the editor's picking behavior is opposite the drawing behavior and USER32's hit test order.
I think that is right and perhaps that is why, as I recall, Visual C went from almost unusable to a pretty good development environment between version 2 and 4. This is all (old) memory and anecdote.
I lost a game prototype to that bug as it crashed my HDD at the same time by writing swap over swap. Screwing up the S.M.A.R.T. page.
Hard reboot left it doa. With today’s tools it could have been recovered but I lacked gparted and a kali thumb stick to bring it back. So thankful for Linux.
Was I the only one that consistently experienced a crash when exiting Visual Studio 6 without first closing the solution? I think this was running on some flavour of Windows 98 in 1999 or 2000. So I got into the habit of closing the solution first.
To this day, I experience a very brief spasm of regret when I just exit Visual Studio without closing the solution first.
The “Visual” in this product’s name was one of the greatest disappointments of my childhood. [1]
I had dabbled in both QuickBASIC and Quick C on MS-DOS building very simple text games and such. When Visual Basic for Windows came out, I was ten years old. It was a revelation because it let me build something that felt like real software. Place controls on a canvas and start wiring them up with event handlers — so easy!
When Visual C++ was announced, I figured it would be the next step up. I had already learned some simple C, so probably it’s just like Visual Basic but you use an improved C for your event handlers? I harangued my dad to “borrow” a copy from his work, and sat down with the Hello World… Which was an incomprehensible 300 lines of Win32 message handlers and Hungarian notation variable names. Nothing “Visual” about it.
It did give me an appreciation for just how much work the Visual Basic team had done to hide the ugly guts of GUI programming.
- -
[1] I had a very easy childhood by most standards.
Nobody who had not done Win16/32 SDK C programming ala Charles Petzold can ever understand what a paradigm shift in programming that Visual C++ with MFC was. I still remember having my boss get the Visual C++ software (around 1995/1996 ?) along with David Kruglinski's two-volume books on programming it, following the tutorial clicking through wizards and being gobsmacked at getting a full-blown ready-to-go app with much of the hard-work done for you in seconds. We rewrote our Win16/32 C app in VC++/MFC and it took us only a few months rather than years.
It was a revelation for me on the power of OOD/OOP in practice and a defining educational moment on Software Design, Frameworks and Libraries.
Those of us coming from Turbo Vision on MS-DOS, OWL on Windows 3.x, and CSet++ on OS/2, saw it as an improvement over Win32, however a worse is better, the worse of all modern C++ GUI frameworks at the time.
Lets also add MFC was born as Application Framework, and dumbed down because hardcore Win16/Win32 C folks at Microsoft were against its original design.
When / How did versioning enter versioning the awkward state we have today? There is cl.exe, MSBuild and build chain at the very least (now at work computer at the moment, pretty sure I am wrong with the making here) with versioning that is close enough to each other to be confusing and related to each other in word ways. Naming itself also feels confusing to me. Documentation also only helps when you already have a good idea what's going on.
>When / How did versioning enter versioning the awkward state we have today? There is cl.exe, MSBuild and build chain
The multiple confusing version numbers are caused by each software component (e.g. "cl.exe" vs "msbuild.exe" vs IDE "devenv.exe" vs platform toolkit SDK) evolving separately over decades. (https://en.wikipedia.org/wiki/Microsoft_Visual_C%2B%2B#Inter...)
E.g. Today's "cl.exe" at version 19.x goes all the way back to 1983 with Microsoft C 1.0 for MS-DOS. Just a text-based command line compiler with no C++, no GUI, no "Studio", etc. 43 years of the core compiler going from 1.x to 19.x. The Visual IDE component of VC++ version 17.x has a different version history going back to 1993 with 1.0. Microsoft C# also has different version numbers of C# language version vs .NET version vs CLR version, etc.
It's analogous to Linux world of different tools with different version numbers. GCC is 15.x, glibc is 2.4, make is 4.4, etc. If you include a "visual GUI IDE" like Jetbrains CLion, that's yet another different version (252.x) that doesn't match any of the others.
Microsoft could hypothetically "synchronize/unify" all version numbers for all disparate products to be a single number ... akin to Apple using version "26" to synchronize Xcode with iOS, etc by introducing large numerical gaps like v18 --> v26. That's probably not going to happen. Likewise, different groups at GNU are not going to agree to synchronize all version numbers such that next release will be GCC 20.0, glibc 20.0, and make 20.0.
Interesting to see the UI change (degradation) over time. The first version makes it very clear what's a button and clickable at first glance, while the latest version looks like a sea of whitespace and makes that far more difficult.
Brings back memories.
But with a focus more on C++ features.
I started with Visual C++ 1.0. There were no templates.
Visual C++ 2.0 had templates.
There was by Visual C++ 3.0. They went to 4.0 to sync the MFC (Microsoft Foundation Classes) versions.
IIRC, you could use a bit of STL with 4.0
Visual C++ 5.0 was mainly optimizer differences.
Visual C++ 6.0 was actually pretty good. However it lacked partial template specialization. I was a Boost author at the time and lots of Boost code had specific workarounds for Visual C++ 6.
Visual C++ 2002 also had no partial template specialization.
Visual C++ 2003 was the first version with partial template specialization and that could compile all of Boost.
Visual C++ 2005 and 2008 did not have much changes.
Visual C++ 2010 tried to get back to 6 in the IDE (there was a deliberate marketing as such). It also had some C++11 features. But no variadic templates.
Visual C++ 2012 had no variadic templates.
Visual C++ 2013 was the first with variadic templates.
Nowadays, Visual C++’is doing much better tracking the C++ standard and often has compiler and library features before clang and GCC.
It doesn't point out the massive productivity boost of Visual C++ 6, IntelliSense (ie, autocomplete). This saved so much time from having to look up documentation.
I worked on that! I was a dev on the team that built the first IntelliSense engine for C++. It’s a miracle it worked at all. It was based on a hacked-up version of the C++ front-end, but when it inevitably hit errors attempting a single-pass parse of the current source, it would silently bull its way through and try not to get too screwed up. Doing that in the presence of templates was not a good time. But with the RAM and processing speeds available then, I’m still kind of shocked it worked at all, 28 years ago.
Oh, very cool! It was like magic when we got that!
No offense but the first thing I used to do after MSVC installation was the installation of Visual Assist :)
Oh man Visual Assist. Was it whole tomato software or something like that? It was amazing back in the day!
nice documentation
heh, sorry, but intellisense was the first thing we (in a game studio) turned off immeadiately after install (by deleting the dll as there was no other way) because it was unacceptable resource hog. this was around VS 2003/2005/2008.
visual assist x worked better.
(Not so) funny that there was nothing "visual" in Visual C++. The dialog resource editor was a joke. I remember how I was relieved when I discovered Borland C++Builder, so I didn't have to switch to Delphi for Windows software development.
Even worse, to this day, Microsoft has completely failed to produce anything comparable to C++ Builder due to the internal feuds around C++ and COM.
C++/CLI could have been it (after the Managed C++ reboot), but its integration with Windows Forms has always been clunky, nevermind WPF.
Then C++/CX could have been it, but then there was this rebellion replacing it with C++/WinRT, nowadays in maintenance, only getting bug fixes, being good enough for WinUI 2.0 and WinAppSDK low level stuff and that is about it.
I was big into Borland from their early MS-DOS compilers, all the way into 1999.
Embarcadero might not be Borland, but still manages to put out great GUI RAD tooling, whereas Microsoft lost sight of what means a great developer experience for native applications.
One thing not mentioned is that Visual C++ and Visual Basic historically were separate IDEs with separate codebases. When the time came to unify them, only one of them could continue on. My understanding is that Visual Basic won, and that today's Visual Studio IDE (devenv.exe, msenv.dll, etc.) is the continuation of that VB codebase.
I don't actually know in which release that transition happened. But since there's a screenshot of each version in the article, presumably that transition is visually documented...
That might have been the transition from Visual C++/Basic 4 to 6 (I skipped 5), but the cataclysmic one was the switch from Visual Studio 6 to Visual Studio .NET, when large portions of the IDE and build system were rewritten in .NET. Visual Studio .NET (2002) was much slower and much buggier than VS6. The native debugger was glacial at conditional breakpoints and debug output, the build system took an eternity to do a dependency check, its UI visibly redrew more slowly, etc. It was so bad that Microsoft had to create a special offer for the Visual Studio .NET (2003) upgrade for only ~$30.
This transition was not great for Visual Basic developers either since their language was transitioned from generating native code (VB6) to becoming dependent upon the .NET Framework (VB.NET), supported secondarily to C#.
To this day I would consider VB 6 the epitome of COM development tooling, neither .NET, nor the UWP trainwreck were as good, and C++ tooling only MFC is worth of mention, as something putting development experience first, instead of Windows division politics.
Quite awkward for what is such a key technology on Windows.
If memory serves, it was Visual InterDev that won, actually. The VID IDE was the forerunner to Visual Studio.NET 2002, which was the first unified Visual Studio IDE.
n.b. I worked on the Visual Studio Core team, which maintained devenv.exe, among other things, from 2003-2007'ish.
Visual InterDev was the first IDE I ever used outside of notepad. It was bundled in with one Microsofts early webediting tools with office (97/200)....but not FrontPage. I thought I was touching the future, it was so amazing! I really don't remember much about other than the impression that it left on me.
After Visual C++ 6. They broke a lot of the C++ IDE features and they weren’t as good as the prior versions (dialog editor etc) so for a long time we preferred staying on 6. I think if we could have the newer compilers but the snappiness of that UI many developers would be happy.
It’s also a product of the segmentation of the developer tools in Microsoft. The Windows team was responsible for the compiler rather than the Developer Tools team.
No, that's not really true. I was on the C++ compiler team from 1991 to 2006. When I first started, the DevTools team reported up through the Windows team, but never really felt a well-integrated part. We were never in the same building as the Windows team, for instance. I remember, probably 1992 or 1993, driving from building 4 where the compiler team lived to the Windows building (forget which one that was, maybe in the 12 to 15 block back then?) to get a copy of the Windows NT source on a hard drive. That's because I was a dev on the C++ compiler back-end team then (moved to the front-end in '95, IIRC), and compiling that source was a major test of the 32-bit compiler I was working on.
Don't remember when DevTools was re-orged out from under Windows, but I'm pretty sure it was by '95, and well before VC++ 6.
Did you ever work on the cross edition that would compile Windows apps for the Mac? I think that was a version 4 fork that never got another version.
No, there were two devs working on the 68k Mac compiler, with ~10 devs on the x86 side (though both targets shared a lot of code and differed mainly in the late codegen and peephole optimization phases). I never worked on the 16-bit code; the 32-bit and later 64-bit x86 backend was a different codebase from the 16-bit stuff.
If I remember from poking around at the 16 bit version in that timeframe (1.52?) there was still an option to generate P-code and (this is now nearly thirty years ago) that was used by Excel to minimize code size for the segmented 16 bit systems.
Any idea how many devs are working at Microsoft on the C and C++ compilers these days? I've heard rumours that there's more on the Rust team and that C++ is taking a back seat.
No clue. I left as a full-time employee in 2007, did a few contractor gigs with various old teams of mine to help out, but that was done by 2013. I lost touch with how things were going internally after that.
The dialog editor is a good example of the damage that was done with the VS.NET transition. When they rewrote it to use the WinForms-based UI, they introduced a fundamental bug: the Z-order for picking was reversed so that clicking on a stack of controls selected the one on the bottom. Very annoying for controls like group controls intended to be stacked below other controls. Bug filed, WONTFIX'd, and it's still broken to this day.
Afaik Win32 API explicitly states that overlapping sibling windows/controls is not supported.
No, overlapping sibling controls is absolutely supported in Win32. It has a defined Z-order for child controls along with functions to manage it:
> The system maintains the z-order in a single list. It adds windows to the z-order based on whether they are topmost windows, top-level windows, or child windows.
https://learn.microsoft.com/en-us/windows/win32/winmsg/windo...
The WS_CLIPSIBLINGS window style is also explicitly for controlling cross-clipping of child windows during painting:
> Clips child windows relative to each other; that is, when a particular child window receives a WM_PAINT message, the WS_CLIPSIBLINGS style clips all other overlapping child windows out of the region of the child window to be updated.
https://learn.microsoft.com/en-us/windows/win32/winmsg/windo...
Furthermore, group boxes and tab controls require overlapping child windows, because they are designed for the controls that visually appear within them to be sibling controls that are raised in Z-order. This is why the dialog editor's current behavior is broken, because the editor's picking behavior is opposite the drawing behavior and USER32's hit test order.
Right! It is wxWidgets that doesn't support overlapping siblings due to other back-ends (besides Win32) lacking support.
the big thing I missed was their macro editor, it was really good for automating things.
I think that is right and perhaps that is why, as I recall, Visual C went from almost unusable to a pretty good development environment between version 2 and 4. This is all (old) memory and anecdote.
You know what this doesn't show?
Visual C++ v4.2b generating an assignment operator which invoked the destructor of a returned object before assigning it to the lval.
That was fun to figure out. :-/
I remember a bug in an old Visual C++ that would fill your RAM and swap (therefore hard drive) if you forgot a semicolon at the end of a class, like:
To fix that, you had to kill the compiler process.I lost a game prototype to that bug as it crashed my HDD at the same time by writing swap over swap. Screwing up the S.M.A.R.T. page.
Hard reboot left it doa. With today’s tools it could have been recovered but I lacked gparted and a kali thumb stick to bring it back. So thankful for Linux.
Nice one... I seem to remember MSVC 5 having a switch that made all functions virtual - or am I confused with another compiler?
Was I the only one that consistently experienced a crash when exiting Visual Studio 6 without first closing the solution? I think this was running on some flavour of Windows 98 in 1999 or 2000. So I got into the habit of closing the solution first.
To this day, I experience a very brief spasm of regret when I just exit Visual Studio without closing the solution first.
The “Visual” in this product’s name was one of the greatest disappointments of my childhood. [1]
I had dabbled in both QuickBASIC and Quick C on MS-DOS building very simple text games and such. When Visual Basic for Windows came out, I was ten years old. It was a revelation because it let me build something that felt like real software. Place controls on a canvas and start wiring them up with event handlers — so easy!
When Visual C++ was announced, I figured it would be the next step up. I had already learned some simple C, so probably it’s just like Visual Basic but you use an improved C for your event handlers? I harangued my dad to “borrow” a copy from his work, and sat down with the Hello World… Which was an incomprehensible 300 lines of Win32 message handlers and Hungarian notation variable names. Nothing “Visual” about it.
It did give me an appreciation for just how much work the Visual Basic team had done to hide the ugly guts of GUI programming.
- -
[1] I had a very easy childhood by most standards.
Nobody who had not done Win16/32 SDK C programming ala Charles Petzold can ever understand what a paradigm shift in programming that Visual C++ with MFC was. I still remember having my boss get the Visual C++ software (around 1995/1996 ?) along with David Kruglinski's two-volume books on programming it, following the tutorial clicking through wizards and being gobsmacked at getting a full-blown ready-to-go app with much of the hard-work done for you in seconds. We rewrote our Win16/32 C app in VC++/MFC and it took us only a few months rather than years.
It was a revelation for me on the power of OOD/OOP in practice and a defining educational moment on Software Design, Frameworks and Libraries.
Those of us coming from Turbo Vision on MS-DOS, OWL on Windows 3.x, and CSet++ on OS/2, saw it as an improvement over Win32, however a worse is better, the worse of all modern C++ GUI frameworks at the time.
Lets also add MFC was born as Application Framework, and dumbed down because hardcore Win16/Win32 C folks at Microsoft were against its original design.
The Windows 2000 era was really peak computing.
Yeah, but not for Microsoft.
When / How did versioning enter versioning the awkward state we have today? There is cl.exe, MSBuild and build chain at the very least (now at work computer at the moment, pretty sure I am wrong with the making here) with versioning that is close enough to each other to be confusing and related to each other in word ways. Naming itself also feels confusing to me. Documentation also only helps when you already have a good idea what's going on.
>When / How did versioning enter versioning the awkward state we have today? There is cl.exe, MSBuild and build chain
The multiple confusing version numbers are caused by each software component (e.g. "cl.exe" vs "msbuild.exe" vs IDE "devenv.exe" vs platform toolkit SDK) evolving separately over decades. (https://en.wikipedia.org/wiki/Microsoft_Visual_C%2B%2B#Inter...)
E.g. Today's "cl.exe" at version 19.x goes all the way back to 1983 with Microsoft C 1.0 for MS-DOS. Just a text-based command line compiler with no C++, no GUI, no "Studio", etc. 43 years of the core compiler going from 1.x to 19.x. The Visual IDE component of VC++ version 17.x has a different version history going back to 1993 with 1.0. Microsoft C# also has different version numbers of C# language version vs .NET version vs CLR version, etc.
It's analogous to Linux world of different tools with different version numbers. GCC is 15.x, glibc is 2.4, make is 4.4, etc. If you include a "visual GUI IDE" like Jetbrains CLion, that's yet another different version (252.x) that doesn't match any of the others.
Microsoft could hypothetically "synchronize/unify" all version numbers for all disparate products to be a single number ... akin to Apple using version "26" to synchronize Xcode with iOS, etc by introducing large numerical gaps like v18 --> v26. That's probably not going to happen. Likewise, different groups at GNU are not going to agree to synchronize all version numbers such that next release will be GCC 20.0, glibc 20.0, and make 20.0.
Last one is 2012... I would like to see a sequel with how it evolved in the last 10 years.
Binders full of MSDN CD-ROMs...IYKYK
[flagged]