• 0

How do you comment?


Question

I know that commenting is a huge debate in programming. I was reading this article on Ars Technica, which is not primarily about commenting, but noticed that the person asking the question noted that he doesn't comment his code because he believes it is self-evident. However, most professional software development firms require commenting of some sort. So what is your personal opinion on commenting your code? Do you comment a lot, not at all, or somewhere inbetween? And, most importantly, why do you take that position?

As I have become a more experienced developer I have developed my own style. By most programmer's standards I believe that I comment quite a bit. I have a couple snipets of my code below, with the full source code for the class attached. (Note: The spacing is slightly screwed up. Not my fault. I blame Neowin's post editor.)


/*
xorangekiller's File System class for easy manipulation of objects on the file system.

This class is essentially a wrapper for Windows API functions.
However, unlike the file manipulation functions in the Windows API, all of these work recursively.

Windows file management functions: http://msdn.microsoft.com/en-us/library/aa364232.aspx
Windows directory management functions: http://msdn.microsoft.com/en-us/library/aa363950.aspx
*/
class xFileSystem
{
public:
// Initialization
xFileSystem();
virtual ~xFileSystem() {};
// Primary File System Utilities
bool CreateDir( const std::string File );
bool Copy( const std::string Source, const std::string Dest );
bool Move( const std::string Source, const std::string Dest );
bool Search( const std::string File, std::string Search = "", bool Case = true, bool Hidden = false, unsigned int Recurse = 0 );
bool Delete( const std::string File );
// Support File System Utilities
static bool Exists( const std::string File );
static bool IsFile( const std::string File );
static bool IsDir( const std::string File );
unsigned long long Size( const std::string File );
[/CODE]

[CODE]
/*
Search for files directories.

Callback:
Every time a matching file or directory is found, SearchCallback() is called.
If it is not overridden, it will save the found files to internal storage, which can be accessed by calling GetFoundFiles() after this function returns.

64-bit Note:
If this function is being run from a 32-bit program on 64-bit Windows you might need to call Wow64DisableWow64FsRedirection() first to allow searching 64-bit paths.
If you use the aforementioned function, don't forget to call Wow64RevertWow64FsRedirection() after it completes to revert the setting!
MSDN Documentation: http://msdn.microsoft.com/en-us/library/aa365743.aspx

Side Effects:
Be advised: If you are using the default implementation of SearchCallback(),
the vector GetFoundFiles() returns is reset (cleared) every time this function is called.

Arguments:
File [in] Directory path to search.
If you supply a file we WILL return with an error!
Search [in] [optional] Find files and directories matching this search string.
Wild characters '*' (any number of characters) and '?' (only one character) are supported.
If the parameter is left blank everything we find will be processed!
Case [in] [optional] Is the search string case sensitive?
This paramter is ignored if Search is "" or "*".
Hidden [in] [optional] Should the search include hidden files and directories?
Recurse [in] [optional] Depth to enumerate the search path.
0 Enumerate the entire path.
1 Enumerate only the base directory.
2 Enumerate the base directory and any directories it contains.
3+ Enumerate all directories up to the designated level.

Return Value:
true We found at least one file or directory matching the search parameters.
flase Nothing was found matching the search parameters.
Are you sure we have permission to access the path?
*/
bool xFileSystem::Search( const std::string File, std::string Search, bool Case, bool Hidden, unsigned int Recurse )
{
//
// Check for trouble.
//
if( !this->IsDir( File ) ) return false; // We can only search directories!
if( Search.empty() ) Search = "*"; // If the search string does not exist, search for all files.
this->FoundFiles.clear(); // Clear our found files vector.
//
// Search the source directory.
//
xDirIt Handle( File, true, Hidden, Recurse ); // Handle for iterating the directory tree.
while( Handle.It() )
{
if( this->Abort ) return false;
// Strip the full directory path.
const char * FileIt = Handle.FileName; // Pointer to the file name that we will iterate through.
const char * LastGood = Handle.FileName; // Last known valid directory path.
while( *FileIt )
{
LastGood = FileIt;
while( *FileIt && *FileIt != '\\' ) FileIt++;
if( *FileIt ) FileIt++; // We don't actually want the '\\' character itself.
};
// Pass on the file information if the file name matches our search criteria.
if( this->WildCmp( Search.c_str(), LastGood, Case ) ) this->SearchCallback( Handle.FileName, Handle.FileAttrib );
};
return true; // Assume that everything ran successfully.
}
[/CODE]

What is your opinion on my commenting? Which category do I fall into, from your perspective.

xfs_1_3_1.zip

Link to comment
https://www.neowin.net/forum/topic/1102831-how-do-you-comment/
Share on other sites

18 answers to this question

Recommended Posts

  • 0

You definitely fall into the "comment a lot" category, but as long as your keep your comments up to date, it's not necessarily a bad thing. It can prove problematic if your comments get out of sync with the actual code though.

Personally I think I sit somewhere in the middle. I believe that your comments should (for the mostpart) explain the "Why" in the code (i.e. why does this code do what it does) and let the code itself say WHAT it does. At the same time though, I think that people who insist on not commenting at all (the "all code should be self-documenting") are dangerous and make life hard for others.

For my code style, I'll generally do the following:

  • Comments at the top of the class, explaining it's purpose.
    /**
     * Unicode string class. Holds a string of unicode characters and provides methods for manipulating the string and conversion utilities.
     */
    class UnicodeString


  • Comments at the top of functions/methods explaining the method, the parameters, the return values, the exceptions returned, and any miscellaneous information relevant to users or maintainers.
    class AsciiString
    {
      /**
       *  Get a ROT-13 encrypted version of this string.
       *
       *  Parameters:
       *    iterations         The number of times to encrypt the string. Should be an odd number.
       *
       *  Return Value:
       *    The ROT-13 encrypted string. If the string is empty, an empty string will also be returned.
       *
       *  Exceptions:
       *    ArgumentException  The number of iterations provided was even
       *
       *  Remarks:
       *    - This method is useless. Seriously.
       */
      String GetRot13(int iterations = 1)
      {
        ..
      }
    }
    


  • TODO and HACK comments. In all my IDEs these are recognised as keywords in comments, and can be useful to refer to at a later date
    int SomeMethod()
    {
      // TODO: Complete this method.
      throw new NotImplementedException("This method is not currently available in the beta version of the API.");
    }
    
    PoorlyImplementedInteger GetSum(PoorlyImplementedInteger values[])
    {
      // HACK: The PoorlyImplementedInteger class does not provide operators to do basic addition, but it does provide comparison
      //       and increment so we can use a loop to do the addition. As soon as addition is implemented update this method to
      //       reflect.
      ...
    }
    


  • Explain the "why".
    string ParseMenuFromTelnetOutput(string commandString)
    {
      // Since telnet is simply just a character stream, we search for the menu input indicator ("$>") and then wait a little
      // longer in case more output is sent.
      ...
    }
    


Those examples are fairly generic, I'll often use documentation-generator syntax when applicable (in C++ I use Doxygen comments, in C# I use intellisense XML, in ruby I use rdoc, etc). I certainly try to avoid commenting every line, since it can simply end up patronising the reader, and confuse them too if the comments get out of sync with the code. Example:

// If index is zero, return false. <--- NO **** SHERLOCK!
if (0 == i) return false;

EDIT: Accidentally a code tag.

  • Like 2
  • 0

Maybe in my example function above my commenting is a bit extreme. I don't normally comment close to every line. (Actually, I didn't realize that function was like that until you pointed it out. There used to be more content there until I abstracted it into a directory iteration class.) I do, however, normally comment every interface (class, struct, function, etc.) and every variable I create. The other comments are optional: they're only there if I decided I need them. My older code tends to have more comments inside function than my newer code does though. Overall, the source I attached to my first post is representative of my commenting and style.

  • 0

Personally I think I sit somewhere in the middle. I believe that your comments should (for the mostpart) explain the "Why" in the code (i.e. why does this code do what it does) and let the code itself say WHAT it does. At the same time though, I think that people who insist on not commenting at all (the "all code should be self-documenting") are dangerous and make life hard for others.

I have another example that may fit what you are talking about, albeit a bit on the extreme side. I created the rather large comment block in this function because I kept confusing myself as to what exactly I need to cleanup in this function and what was taken care of by the thread it spawns. The system as a whole that this code is part of is rather complex. This may be the single largest comment block I have ever written inside of a function.


/*
Load the specified plugin.

Remarks:
WARNING: This function has high overhead!

Arguments:
File [in] Name and path of the plugin to load.

Return Value:
If the plugin was loaded the identification number of the thread will be returned.
If the plugin path is invalid or the plugin has already been loaded 0 will be returned.
*/
unsigned long PluginManager::LoadPlugin( const char * File )
{
if( File == NULL ) return 0;
if( this->IsPluginLoaded( File ) == true ) return 0;
this->Plugins.push_back( PluginThread() );
this->Plugins.back().File = File;
this->Plugins.back().Provider = new EVPlugin( this->ProviderBase, 0, this->ED, this->Tabs );
this->Plugins.back().Provider->InfoMsg() << "Loading plugin " << File << std::endl;
// Before we actually create the thread, here is a short analysis of how the members of this->Plugins.back() are used.
//
// File File name and path of the plugin to load.
// This is also used by us as a sentinel to tell when the thread is finished initializing.
// Since the file name is no longer needed after the plugin is loaded (and it is not guaranteed
// to be initialized after this function exits), NULL is used as the sentinel value.
//
// Provider The plugin service provider handle that allows the plugin to interface with Alekto.
// It must be initialized in this function because only PluginManager can access the base.
// However, it does not need to be deleted anywhere in this class.
// Since it is used by the plugin thread exclusively, it will be deleted before the thread exits.
//
// Plugin The loaded plugin's handle.
// It is both created by and destroyed by the thread.
//
// Handle The handle of the plugin's thread.
// This handle must be assigned in this function when the thread is created.
// However, since it is no longer valid after the thread exits, the thread is responsible for nullifying it.
// This nullification is not done until the thread is ready to exit.
// Therefore, it may be safely used as a sentinel for thread exit.
this->Plugins.back().Handle = CreateThread( NULL, 0, (LPTHREAD_START_ROUTINE) &PluginThreadProc, reinterpret_cast< void * >( &this->Plugins.back() ), 0, NULL );
while( this->Plugins.back().File != NULL ) Sleep( 1000 ); // Wait until the thread is done loading the plugin.
Sleep( 500 ); // Wait for another half-second, just to be sure the thread has had sufficient time to cleanup.
if( this->Plugins.back().Handle == NULL )
{
this->Plugins.pop_back();
return 0;
};
return EVPlugin::EncodeID( this->Plugins.back().Handle );
}
[/CODE]

  • 0

I have another example that may fit what you are talking about, albeit a bit on the extreme side. I created the rather large comment block in this function because I kept confusing myself as to what exactly I need to cleanup in this function and what was taken care of by the thread it spawns. The system as a whole that this code is part of is rather complex. This may be the single largest comment block I have ever written inside of a function.

FWIW, when I said "your" in that quote I was speaking generalities, it wasn't intended as a commentary on your commenting style :p

  • 0

People that say they don't comment because it's self evident are usually the ones who don't know what they're really doing and cause a massive headache for others once they are fired/leave and we're left with a jumbled mess of spaghetti code.

  • Like 2
  • 0

Well in .net I use the built-in XML style commenting system - a quick example:


/// <summary>
/// An example method to show 'Commenting Style'.
/// Performs no real functionality.
/// </summary>
/// <param name="FirstName">String - Input Persons Forename</param>
/// <param name="LastName">String - Input Persons Surname</param>
/// <param name="ID">Short Integer - Input Persons ID Number</param>
public void Example(string FirstName, string LastName, Int16 ID)
{
// Blah Blah Code Here
}
[/CODE]

  • 0

I comment more than I probably should, my code is clean and concise, but I like to outline my code in pseudo code before I write it, then I write the code to match the comments. So you could either read the comments to know what's happening, or read the code.

  • 0

I don't put comments in my code unless it's code I borrowed from some open source app, or something I got from a StackOverflow question. Then I will put what app it came from or the link to the SO question.

I also don't code for a living, nor do I do anything that is overly complex enough to actually require comments for someone to figure out what it does. I try to use function and variable names that would make it obvious to a reader what a particular block of code does. Same for class names.

  • 0

IMO I think you put in too many comments in. Code in my option should be simple and mostly self documenting. Keep It Simple and clean. Too much documentation can make it difficult to read actual code.

as some one said

I try to use function and variable names that would make it obvious to a reader what a particular block of code does. Same for class names.

return true; // Assume that everything ran successfully.

do you need to add that comment in ? ESPECIALLY when you have documented what it returns in the method header....

if you used a variable like 'hasRunSuccessfully' and then have 'return hasRunSuccessfully' it is essentially self documenting - no need for comments.

//

// Check for trouble.

//

check for 'trouble' not the best label - but programmers dont have to use the best labels, no need for the extra // just put in a blank line.

this->FoundFiles.clear(); // Clear our found files vector.

I would label this as FoundFileList or FoundFilesVector or FoundFileCollection and then if you have FoundFilesVector.Clear() it is pretty self evident what you are doing.

  • 0

IMO it boils down to adding relevant information where it's likely to help. There's no universal rule to determine what information is relevant and when it's likely to help. Rules like comment every function, or don't comment x or y, don't work. I am particularly vexed by any mandatory ASCII artwork like drawing boxes and lining up stars in front of every function. Complete waste of my time.

I think a combination of a clean, self-documenting coding style and unit testing goes a long way towards providing good understanding and confidence about a code base; unlike comments they can't lie and they can't go out of date unnoticed.

  • 0

I comment only where something is not clear and where the IDE makes use of them.

Our team follows the principals of self-explaniatory code. It's must easier to read a small function that does one thing with a set of self explanatory actions than a large function loaded with comments.

We'd probably write something closer to this for the example provided above:


unsigned long PluginManager::LoadPlugin( const char * fileName ) {

  if (!this-&gt;ShouldLoadPlugin( fileName ))
    return 0;

  PluginThread* pluginThread = this-&gt;CreatePluginThread( fileName );

  this-&gt;LoadPluginAndWait( pluginThread );

  return this-&gt;GetPluginResult( pluginThread );

}

  • Like 1
  • 0

By the way, the comments are the least of your worries about the last function you posted. In order of importance:

1)


while( this->Plugins.back().File != NULL ) Sleep( 1000 ); // Wait until the thread is done loading the plugin.
Sleep( 500 ); // Wait for another half-second, just to be sure the thread has had sufficient time to cleanup.[/CODE]

Waiting half a second doesn't guarantee any particular thread will run at all. This might come up 1 time out of 10000 in particular conditions which you might not even be able to reproduce. You should be using a proper thread synchronization mechanism such as a monitor here.

2)

[CODE] this->Plugins.back().Provider = new EVPlugin( this->ProviderBase, 0, this->ED, this->Tabs );[/CODE]

Any "new" operator in methods is a red flag. Exceptions can be thrown anywhere (including by the "new" operator itself) and prevent any cleanup code from executing, leading to memory leaks. Plus since there's no corresponding "delete" within the method, the cleanup responsibility is apparently held by some different code, which makes your memory management difficult to reason about (and probably wrong). The only way to guarantee proper cleanup in C++ is RAII, use it.

3) Some stylistic issues:

- Inconsistent use of C/C++-style casts:

[CODE]CreateThread( NULL, 0, (LPTHREAD_START_ROUTINE) &PluginThreadProc, reinterpret_cast< void * >( &this->Plugins.back() ), 0, NULL );[/CODE]

- Using "this->" in front of every member access is just syntactic noise.

- If you're using a recent C++ compiler (VS2010 and above), use nullptr instead of NULL. nullptr is a pointer type, NULL is an integral constant. Consider:

[CODE]
void foo(int*);
void foo (int);

void bar() {
foo (NULL); // Calls 'foo(int)'
}[/CODE]

  • 0

By the way, the comments are the least of your worries about the last function you posted. In order of importance:

1)


while( this->Plugins.back().File != NULL ) Sleep( 1000 ); // Wait until the thread is done loading the plugin.
Sleep( 500 ); // Wait for another half-second, just to be sure the thread has had sufficient time to cleanup.[/CODE]

Waiting half a second doesn't guarantee any particular thread will run at all. This might come up 1 time out of 10000 in particular conditions which you might not even be able to reproduce. You should be using a proper thread synchronization mechanism such as a monitor here.

2)

[CODE] this->Plugins.back().Provider = new EVPlugin( this->ProviderBase, 0, this->ED, this->Tabs );[/CODE]

Any "new" operator in methods is a red flag. Exceptions can be thrown anywhere (including by the "new" operator itself) and prevent any cleanup code from executing, leading to memory leaks. Plus since there's no corresponding "delete" within the method, the cleanup responsibility is apparently held by some different code, which makes your memory management difficult to reason about (and probably wrong). The only way to guarantee proper cleanup in C++ is RAII, use it.

3) Some stylistic issues:

- Inconsistent use of C/C++-style casts:

[CODE]CreateThread( NULL, 0, (LPTHREAD_START_ROUTINE) &PluginThreadProc, reinterpret_cast< void * >( &this->Plugins.back() ), 0, NULL );[/CODE]

- Using "this->" in front of every member access is just syntactic noise.

- If you're using a recent C++ compiler (VS2010 and above), use nullptr instead of NULL. nullptr is a pointer type, NULL is an integral constant. Consider:

[CODE]
void foo(int*);
void foo (int);

void bar() {
foo (NULL); // Calls 'foo(int)'
}[/CODE]

I really appreciate your feedback. Especially RAII, which I didn't know about before.

I do have one question though. I am using GCC 4.6.2 to compile the project the code excerpt you quoted in point 3 came from, so it does have (mostly complete) C++11 support and I can use nullptr. With the size of this project, however, it would take considerable effort to convert all of my null pointers to nullptr instead of NULL. What is the advantage of doing so?

  • 0

I do have one question though. I am using GCC 4.6.2 to compile the project the code excerpt you quoted in point 3 came from, so it does have (mostly complete) C++11 support and I can use nullptr. With the size of this project, however, it would take considerable effort to convert all of my null pointers to nullptr instead of NULL. What is the advantage of doing so?

You could just run find-and-replace over the entire project, if you're feeling bold. It should just work, in theory. The benefit of nullptr over NULL is in certain cases where there is ambiguity between integer and pointer type, where using NULL can result in unexpected bugs. nullptr is a proper pointer type unlike NULL, so it behaves as expected.
  • 0

I try to comment whenever possible in my code, block comments above classes / methods and small lines for clarification where it may be needed.

I find this method makes it easier when you get stuck, and helps identify logic errors more easily. Not only does it help to maintain the code but it helps me think through a de-bugging process.

If I know somebody else will be reading / maintaining / using my code I make sure I have legible nearly organised comments.

This topic is now closed to further replies.
  • Posts

    • Win11Debloat 2026.06.14 by Razvan Serea Win11Debloat is a lightweight, easy to use PowerShell script that allows you to quickly declutter and customize your Windows experience. It can remove pre-installed bloatware apps, disable telemetry, remove intrusive interface elements and much more. The script also includes many features that system administrators and power users will enjoy. Such as a powerful command-line interface, support for Windows Audit mode and the option to make changes to other Windows users. All changes made by Win11Debloat can be easily reversed, and most removed apps can be restored via the Microsoft Store. A full guide on how to undo the changes is available here. Win11Debloat features: Below is an overview of the key features and functionality offered by Win11Debloat. Please refer to the wiki for more information about the default settings preset. Remove a wide variety of preinstalled apps. Click here for more info. Disable telemetry, diagnostic data, activity history, app-launch tracking & targeted ads. Disable tips, tricks, suggestions & ads across Windows. Disable Windows location services & app location access. Disable Find My Device location tracking. Disable 'Windows Spotlight' and tips & tricks on the lock screen. Disable 'Windows Spotlight' desktop background option. Disable ads, suggestions and the MSN news feed in Microsoft Edge. Hide Microsoft 365 ads on the Settings 'Home' page, or hide the 'Home' page entirely. Disable & remove Microsoft Copilot. Disable Windows Recall. Disable Click to Do, AI text & image analysis tool. Prevent AI service (WSAIFabricSvc) from starting automatically. Disable AI Features in Edge. Disable AI Features in Paint. Disable AI Features in Notepad. Disable the Drag Tray for sharing & moving files. Restore the old Windows 10 style context menu. Turn off Enhance Pointer Precision, also known as mouse acceleration. Disable the Sticky Keys keyboard shortcut. Disable Storage Sense automatic disk cleanup. Disable fast start-up to ensure a full shutdown. ...and more. Once you’ve downloaded the Win11Debloat file (Get.ps1), just follow these quick steps: Locate the Get.ps1 script file. Right-click the file and select Run with PowerShell from the context menu. If prompted by User Account Control (UAC), select Yes to grant the script the necessary administrative permissions. Win11Debloat 2026.06.14 changes: This is a minor release that hopefully addresses the false positives in Windows Defender and Bitdefender that prevented users from downloading and/or running Win11Debloat. Refactor Get-RegFileOperations.ps1 to address false positives by @Raphire in #626 Add logging around WinGet app retrieval and increase timeout to 20s by @Raphire Download: Win11Debloat 2026.06.14 | Open Source View: Win11Debloat Home Page | Screenshots 1| 2 Get alerted to all of our Software updates on Twitter at @NeowinSoftware
    • Still using Microsoft Money 2005 in 2026 here!
    • I have a couple to mention, and they still run great on Windows 11 Adobe Lightroom Version 2 Alcohol 120% CLZ Book, Comic, Game, Movie, & Music Collector (PC - No longer sold / Grandfathered in - now mobile apps/online only) DVDDecrypter ISO Buster Pro version 1.9.1 (Still supports HD-DVD too) Nero Burning Rom 8 (Only the burning software, no backup, media converter, etc)   OpenAL (Runtime) - GuildWars 1 Reforged still uses it for 3d headphone audio PowerDVD 12 Ultra SPTD (SCSI Pass through Direct Driver) UltraISO Windows Media Encoder 9 WinImage You can tell I still sport an optical drive    
    • Linux 7.1 arrives with an NTFS overhaul and major hardware performance boosts by Paul Hill The founder of the Linux kernel has just announced the availability of Linux 7.1. This is a stable version of the kernel that will now be tested by various Linux distributions before it is shipped to users through update managers. Some users, like those on Debian, for example, might not get it for a long time, if at all, while Fedora users can expect it in the near future. With Linux 7.1 out on time, the merge window for Linux 7.2 is now open, giving contributors the opportunity to send in major new features that have been waiting for the last two months. Torvalds warned that he is currently travelling and will be in another timezone, so timing for the merge window may be irregular due to timezone differences and limited internet access. Torvalds said that he has already fetched early pull requests to allow him to do some offline work, but the travel could still cause disruption. Right now, he is not planning to extend the release, but did consider it. He said he might later regret not extending, though. In terms of this last week of development for Linux 7.1, Torvalds said there were no major or alarming changes. This week consisted mostly of smaller driver updates to GPU, networking, and sound, networking fixes, trace tooling fixes, and misc minor fixes. The shortlog this week lists fixes for driver bugs, memory leaks, I/O and USB fixes, networking and RDMA fixes, DRM/graphics fixes, and tooling and verification improvements. Specific fixes include USB series heap-overflow and buffer overflow fixes, and multiple use-after-free, memory-leak, and refcount corrections across subsystems such as i2c, zram, gpio, and net. There are fixes for graphics drivers, including amdgpu, i915, and virtio, as well as hypervisor and virtualization tweaks affecting mshv, vmbus, and hyperv. According to Phoronix, anyone running Linux 7.1 should look out for the new NTFS driver, Intel FRED for improved performance on Panther Lake and future CPUs, faster graphics with Intel Arc Battlemage, and improvements for older AMD Radeon GPUs. If you are running Linux on your computer and everything is fine, then you don’t need to worry about updating to Linux 7.1 as a priority; just wait for it to be pushed to you. If you have tried Linux on hardware but it didn’t work properly, trying again with a distro that uses Linux 7.1 could cause Linux to work on your machine, thanks to the new hardware support.
    • you can also do this with this tool: PowerSettingsExplorer made by mbk1969 at 3dguru forum.. I found it by accident researching on modern standby and annoying quirks of it in 2022
  • Recent Achievements

    • Week One Done
      rolfus earned a badge
      Week One Done
    • One Month Later
      Leroy Jethro Gibbs earned a badge
      One Month Later
    • Conversation Starter
      flexorcist earned a badge
      Conversation Starter
    • One Month Later
      AndreaB earned a badge
      One Month Later
    • One Month Later
      agatameier earned a badge
      One Month Later
  • Popular Contributors

    1. 1
      +primortal
      507
    2. 2
      +Edouard
      197
    3. 3
      PsYcHoKiLLa
      139
    4. 4
      ATLien_0
      90
    5. 5
      Steven P.
      81
  • Tell a friend

    Love Neowin? Tell a friend!