Musings on extended stored procedures and the cloud

(The term “extended stored procedure” as referred to herein is shorthand for the general technique, available in many modern RDBMSes including Microsoft SQL Server, Oracly, MySQL et al., of linking in compiled code written in non-SQL languages for invocation as stored procedures, functions, and the like.)

I recently ran into a newfangled version of DLL hell (an example support post; there are many). The basic issue was that in a SQL Server CLR stored procedure of ours, the production host machines had a different version of some .NET assemblies in the GAC, or Global Assembly Cache. It was a platform assembly causing the issue, in this case resolved by running the following statement in each affected database:

ALTER ASSEMBLY [System.Runtime.Serialization] FROM 'C:\Windows\Microsoft.NET\Framework64\v4.0.30319\System.Runtime.Serialization.dll'
WITH PERMISSION_SET = UNSAFE;

 

While prompted by security measures in Microsoft’s solution, this is not a very elegant or maintenance-free approach. Imagine a future patch bringing down production systems due to a newer version of a DLL suddenly included in GAC, for instance. The cloud makes things even more complex, since having to include even platform DLLs with lowered-safety permissions may make an application difficult or impossible to deploy in a cloud implementation such as Amazon’s RDS.

Unfortunately, there are some highly useful third-party libraries which may in a naive implementation require such extra problematic configuration, such as JSON.NET . Due to understandable security concerns with linking in unvetted code, most modern RDBMSes supporting extended stored procedures will make one jump through extra hoops to use any code which makes system-level calls, can perform I/O, and the like.

Problems like this can be avoided (while still reaping the benefits of stored procedures) with approaches like the following, each with pros and cons:

1. Avoid linking in general utility libraries with references to unsafe code.

2. Create a safe version of reusable utility code for inclusion in heightended-security containers and environments. Store this in a separate project or other compilable unit from the unsafe reusable code.

3. Investigate the use of data-platform features instead of potentially unsafe program code. For example, SQL Server 2016 includes certain features for working with JSON data that might make a library like JSON.NET largely unnecessary in a particular situation.

4. Find safer alternatives to popular libraries. To tack on to the JSON example, small, fast JSON parsers could avoid the need for something like JSON.NET, and even give performance enhancements to boot.

5. If all else fails, consider importing an unsafe library into your codebase and stripping out unsafe calls. This has the obvious drawback of increased code maintenance in case of any necessary patches, but could work for a stable, mature library. Freezing a version in this manner might actually improve the long-term stability of a dependent codebase in some cases.

Of course, this is all really another facet of planning for the cloud. In my current organization, we’re shifting away from using proprietary software like SQL Server in favor of open-source alternatives like MySQL, MariaDB, and MongDB partly for licensing reasons, but just as much for the benefits of broad cross-cloud support. Growing pains for companies shifting more systems to the cloud can be minimized with a shared approach, which explicitly plans a set of reusable, cloud-safe libraries from the start.

Advertisements

Discover all non-abstract types implementing an interface in C#

It’s sometimes useful to auto-discover types in C# and other high-level programming languages, even outside the context of coding an IoC container. In C#, robust type-discovery mechanisms make this easy. Here’s a basic formula for auto-discovery of types implementing an interface:

1. Get the list of assemblies in the application domain.
2. Get the types from each discovered assembly.
3. Test each type to see whether it implements the target interface.

This basic formula can be accomplished in a snap using Linq:

var targetType = typeof(someInterface);
var types =
    AppDomain.CurrentDomain.GetAssemblies()
    .SelectMany(a => a.GetTypes())
    .Where(t => t.GetInterfaces().Contains(targetType) && t.IsClass && !t.IsAbstract)
    .OrderBy(t => t.FullName);

Note that t.IsClass (or !t.IsInterface) in the above expression excludes the interface itself, as well as derived interfaces. The .IsAbstract check can be skipped if the goal is to discover all class types for the interface, not just those that can actually be instantiated. Lastly, the basic approach is similar but the calls not identical for .NET Core.

Combine approaches for easy, lightweight JavaScript user information detection

Basic browser/user detection can be important for providing an enhanced user experience. For example, making a best-guess detection of the user’s IP address (subject to the usual caveats about VPN use) can help in pre-selecting a user’s country of origin, enabling faster form fill-outs. The user’s preferred language, especially on first access of a site, is obviously of huge importance; detecting it early with a non-intrusive script can help avoid clunkier approaches such as passing out language-differentiated URLs to globalized versions of your site, which may be passed between users with unintended consequences.

Especially for a site which already makes use of JavaScript, an unobtrusive script can make a difference. And while there are fairly robust libraries such as platform.js, they often do not address all concerns of detecting language, geoIP, and browser. Adding extra dependencies to a site, in terms of maintenance burden and JS payload, is desirable to avoid when possible.

It’s easy enough to create a tiny script which merges freely available geoIP with basic browser/language detection, as shown below. There are many free geoIP web services available these days; a first implementation can simply rely on one, perhaps adding a fallback as necessary, though eventually a more robust solution may become necessary. (Edit: The freegeoip.net service will become paid-only in July 2018.)

The simple demo code, with browser and OS versions not yet filled in, appears below, and may be run at JSFiddle . This could, for example, be passed as a JSON bundle in a call to a web service, or submitted with a page. While the code may be reused freely, as it essentially cobbles together multiple simple well-known methods into an easy-to-use package, the emphasis is on the approach: a lightweight script that avoids heavy dependencies and still captures useful information about a user. As always, use such information responsibly, especially keeping in mind GDPR regulations which will have taken full effect by May 2018.

function UserInfo() {
  var self = this;
  this.language = ((navigator.languages ? navigator.languages[0] : (navigator.language || navigator.userLanguage)) + '').substring(0, 2);
  this.browserName = '';
  this.browserVersion = '';
  this.osName = '';
  this.osVersion = '';
  this.country = '';
  this.ipAddress = '';
  this.city = '';
  this.region = '';
  this.postalCode = '';
  
  if (navigator && navigator.userAgent) {
    var ua = (navigator.userAgent + '').toLowerCase();
    var platform = ua;
    
    var i = platform.indexOf('(');
    if (i > -1)
      platform = platform.substring(i + 1);

    i = platform.indexOf(')');
    if (i > -1)
      platform = platform.substring(0, i);
      
    if (platform.indexOf('windows') > -1)
      this.osName = 'Windows';
    else if (platform.indexOf('android') > -1)
      this.osName = 'Android';
    else if (platform.indexOf('macintosh') > -1 || platform.indexOf('macos') > -1)
      this.osName = 'macOS';
    else if (platform.indexOf('iphone') > -1 || platform.indexOf('ipad') > -1)
      this.osName = 'iOS';
  }

  if (!!window.chrome && !!window.chrome.webstore)
    this.browserName = 'Chrome';
  else if (typeof InstallTrigger !== 'undefined')
    this.browserName = 'Firefox';
  else if (/*@cc_on!@*/false || !!document.documentMode)
    this.browserName = 'IE';
  else if (/constructor/i.test(window.HTMLElement) || (function (p) { return p.toString() === "[object SafariRemoteNotification]"; })(!window['safari'] || safari.pushNotification))
    this.browserName = 'Safari';
  else if ((!!window.opr && !!opr.addons) || !!window.opera || navigator.userAgent.indexOf(' OPR/') >= 0)
    this.browserName = 'Opera';
  else if (!!window.StyleMedia)
    this.browserName = 'Edge';
  
try {
    $.getJSON('//freegeoip.net/json/?callback=?', function(data) {
      self.ipAddress = data.ip + '';
      self.city = data.city + '';
      self.region = data.region_code + '';
      self.country = data.country_code + '';      
      self.postalCode = data.zip_code + '';
    });
} catch (err) {}

  this.ids = []; 
  
  this.addID = function(sn, st, idt, i) {
    var newID = {};
    newID.systemName = sn + '';
    newID.systemType = st + '';
    newID.idType = idt + '';
    newID.id = i + '';      
    this.ids.push(newID);
  };  
}

 

The minified version is relatively compact, at 1560 characters:

function UserInfo(){var i=this;if(this.language=((navigator.languages?navigator.languages[0]:navigator.language||navigator.userLanguage)+"").substring(0,2),this.browserName="",this.browserVersion="",this.osName="",this.osVersion="",this.country="",this.ipAddress="",this.city="",this.region="",this.postalCode="",navigator&&navigator.userAgent){var e=(navigator.userAgent+"").toLowerCase(),o=e.indexOf("(");o>-1&&(e=e.substring(o+1)),(o=e.indexOf(")"))>-1&&(e=e.substring(0,o)),e.indexOf("windows")>-1?this.osName="Windows":e.indexOf("anddroid")>-1?this.osName="Android":e.indexOf("macintosh")>-1||e.indexOf("macos")>-1?this.osName="macOS":(e.indexOf("iphone")>-1||e.indexOf("ipad")>-1)&&(this.osName="iOS")}window.chrome&&window.chrome.webstore?this.browserName="Chrome":"undefined"!=typeof InstallTrigger?this.browserName="Firefox":document.documentMode?this.browserName="IE":/constructor/i.test(window.HTMLElement)||"[object SafariRemoteNotification]"===(!window.safari||safari.pushNotification).toString()?this.browserName="Safari":window.opr&&opr.addons||window.opera||navigator.userAgent.indexOf(" OPR/")>=0?this.browserName="Opera":window.StyleMedia?this.browserName="Edge":(isChrome||isOpera)&&window.CSS&&(this.browserName="Blink");try{$.getJSON("//freegeoip.net/json/?callback=?",function(e){i.ipAddress=e.ip+"",i.city=e.city+"",i.region=e.region_code+"",i.country=e.country_code+"",i.postalCode=e.zip_code+""})}catch(i){}this.ids=[],this.addID=function(i,e,o,s){var t={};t.systemName=i+"",t.systemType=e+"",t.idType=o+"",t.id=s+"",this.ids.push(t)}}

 

Selecting unthemed HTML descendant elements using CSS :not()

In my recent work, I’ve been constructing a CSS framework that needs to support multiple themes: the ability to override many aspects of display formatting by applying a CSS class to a parent element, affecting all descendants. However, in order to make this work properly in actual practice, it’s may be desirable to apply default theming rules to even children without an ancestor theme-tagged element. (In my situation, this requirement applies because the mentioned CSS framework will be used to support microsites, which should be presented with a “good” look and feel even if the implementation of the site has been somewhat sloppy.)

The CSS :not() selector works well for this, as implemented in CSS3:

:not([class*=Theme]) * {
  // Sample selecting all unthemed elements, where themes are applied with *Theme class names on ancestors
}

This can easily, of course, be applied with different descendant selectors. Try the example.

Easily execute dynamic C# using extension methods

Code has now been released under the SharpByte project to execute dynamic C# scripts (and evaluate statements) more easily than ever before. Dynamic code execution during earlier days of .NET was a sore spot, with many lamenting the lack of a functional equivalent to the JavaScript eval() function. For years many developers attempted hacks like using ASP.NET’s DataBinder.Eval(), but results were often subpar and performance was lackluster. Compiling to the CodeDom and the newer .NET Compiler Platform, a.k.a. Roslyn, can be moderately simple to complex depending on need, but many developers just want a simple, easy-to-reuse solution for supporting dynamic code entry in an application.

Further documentation on the easy-to-use API will be forthcoming, but for now these steps will suffice for anyone wishing to play with the code:

1. Either build and reference the project’s core assembly in your project, or import the code directly into your project.

2. If the code was built with conditional compilation symbol GLOBAL_EXTENSIONS, all objects will be able to use the dynamic-execution extension methods. Otherwise, if COLOCATE_EXTENSIONS was used, add a using statement for the System.Runtime.CompilerServices namespace; if neither GLOBAL_EXTENSIONS nor COLOCATE_EXTENSIONS was used, add a using statement for the SharpByte.Dynamic namespace.

3. Call any version of .Execute() or .Evaluate() directly on any object. The former will execute any C#-compliant script composed of properly semicolon-terminated lines of code, with “this” references executed on the object on which the extension methods are called (i.e. the context object for the call); the latter will evaluate a C# expression and return the result.

Once these steps are done, calling a script is as easy as this:

someObject.Execute("[script statements go here, and may be many lines]");

To anyone curious enough to understand the working of these extension methods, the code will be illustrative. Essentially, the extension methods call into a facade for compilation features of the .NET framework, and can be used to front-end calls to the CodeDom, the .NET Compiler Platform (“Roslyn”), or any other compiler, vastly simplifying the most-needed dynamic code compilation and execution features of each.

Here’s an example of the relatively complex task of building code using the CodeDom (without any attempt to slam the useful-looking code at this page, just provide an example of the complexity hidden away). Roslyn provides many enhancements over CodeDom, but still to simply execute a script, such as user-entered code in a CMS or other data system, isn’t always completely simplified as it could be.

The provided reference code compiles code constructed on-the-fly using the referenced compiler. A code formatter emits source code, without needing any special knowledge of the underlying compiler. A quick review of the System.Object extension methods involved shows how easy it is to retain a reference to the compiled IExecutable instance as well, which can be used to inspect the built-in execution log and timings, as well as any exception generated by the last run of a compiled executable. A unique signature based on the executable code type (expression/statement or script), parameter names, and source code is used on subsequent calls to check for pre-existing compiled executables, stored in the ExecutableFactory hybrid factory/collection for reuse.

Each executable can be compiled successfully with numbered placeholder values, a la string.Format() (but using triple curly braces to avoid .NET and Handlebars-style format interference) and/or named parameter values. As mentioned above, the context object (when using extension methods, the object on which the method is called) is used for any references to “this” within any script or expression.

Since an object from each compiled executable type has a Copy() method, it can safely and cheaply be used to create further executables of the same type. Calling Execute() on any particular executable is guaranteed to be thread-safe due to use of synchronization; for that reason, it’s easiest to cache local copies of reusable expressions/scripts.

Performance-wise, on a fairly low-spec laptop, formatting source code for a new class tests in the 1-2 microsecond range; post-compilation, execution of a script or expression can take well under a microsecond (e.g. a two-parameter complex mathematical expression which tests in the 800-nanosecond range), depending of course on complexity. The bulk of this fairly small overhead is due to the use of dynamic variables within the compiled classes themselves. If warranted, type safety may be added to the context object and/or named parameters to boost peformance further.

This generic utility code was originally developed in support of the SharpByte CMS, but is provided separately under the MIT License. Happy coding!

Safely navigating object hierarchies in JavaScript using prototype methods

New: Dynamically evaluate C# expressions and execute C# scripts with a single statement, from anywhere in a .NET application. Click here for more info.

Anyone who’s dealt with a deeply nested set of properties in JavaScript, whether through use of an extensive third-party JavaScript API or a custom library, has likely run into the problem of safely accessing such structures. One can of course hand-roll many ugly, hard-to-read statements like the following, adding on to the maintenance burden of the code (splitting lines as necessary for length, of course):

if (a && a.b && a.b.c && a.b.c.d && a.b.c.d.e) { doSomethingWith(a.b.c.d.e); }

The main alternative to this approach is to make such accesses within a try-catch block, but this is generally slower by at least a couple of orders of magnitude when exceptions are thrown, so not always useful in tight loops and other performance-sensitive situations. It’s also arguably an abuse of the try/catch mechanism.

Luckily, a less unsavory solution with fairly good performance can be adopted using JavaScript prototype methods. Here’s a reference implementation, and you can also try it for yourself (with timings):

// Indicates whether an object has the indicated nested subproperty, which may be specified with chained dot notation 
// or as separate string arguments.
Object.prototype.hasSubproperty = function() {
	if (arguments.length == 0 || typeof(arguments[0]) != 'string') return false;  
  var properties = arguments[0].indexOf('.') > -1 ? arguments[0].split('.') : arguments;    
  var current = this;
  for(var x = 0; x  -1 ? arguments[0].split('.') : arguments;    
  var current = this;
  for(var x = 0; x < properties.length; x++) {
  	current = current[properties[x]];
    if ((typeof current) == 'undefined') return undefined;
  }  
  return current;
};

// Gets the indicated nested subproperty, which may be specified with chained dot notation or as separate arguments.
// If the specified subproperty (or any intervening object in the hierarchy) is not found, returns undefined.
Object.prototype.getSubproperty = function() {
	if (arguments.length == 0 || typeof(arguments[0]) != 'string') return false;  
  var properties = arguments[0].indexOf('.') > -1 ? arguments[0].split('.') : arguments;    
  var current = this;
  for(var x = 0; x < properties.length; x++) {
  	current = current[properties[x]];
    if ((typeof current) == 'undefined') return undefined;
  }  
  return current;
};

// Sets the indicated nested subproperty, which may be specified with chained dot notation or as separate arguments.
// If any intervening object in the hierarchy is not found, returns false, otherwise sets the value and returns true.
Object.prototype.setSubproperty = function() {
	if (arguments.length  -1 ? arguments[0].split('.') : Array.prototype.slice.call(arguments, 0, arguments.length - 1);    
  var parent, current = this;
  for(var x = 0; x < properties.length - 1; x++) {
  	current = current[properties[x]];
    if ((typeof current) == 'undefined') return false;
  }  
  current[properties[properties.length - 1]] = arguments[arguments.length - 1];
  return true;
};

Some observations: if you run the timings, you’ll note that the try-catch method is still quite fast when exceptions are not thrown, indicating that try-catch might be workable when exceptions are expected to be truly… exceptional. Still, in any but extraordinary conditions, the performance of the prototype-method approach should be quite fast enough, avoids worst-case performance, and is cleanest overall.

A C# queue package designed for high performance and ease of use

New: Dynamically evaluate C# expressions and execute C# scripts with a single statement, from anywhere in a .NET application. Click here for more info.

I’ve recently published a carefully constructed set of generic queues as part of the SharpByte project. These were all developed pursuant to the SharpByte CMS (the first of its type–more about which will be forthcoming), but are publicly released under the MIT license for any who wish to use them.

The priority queue abstract data type is similar to a regular queue, but allows each item to have a priority associated with it. For certain types of server-side programming this can be extremely useful for controlling the flow of work items. Implementations in the package include PriorityQueue and ConcurrentPriorityQueue.

The double-ended queue can be just as useful in its own way, as it can function as a queue, list (if list semantics are supported, as they are with this implementation), or stack. As such it’s one of the most useful general-purpose data structures. Implementations in the package include DoubleEndedQueue and ConcurrentDoubleEndedQueue.

Each queue in the package supports full list operations, and in fact implements IList and IList<T>. Performance has been tested and throughput on the queues under load was seen to be fast under a range of different processing conditions, including multithreading.

All queues in the package can also serve as connectors or work routers within an application, as they can have one or more outputs added, which can be any .NET collection capable of receiving items of the generic type supported by a particular queue. The semantics for this can be seen in the IQueue interface. Strategies for distribution to child queues/collections include multicasting, simple load-balancing, and round-robin distribution.

Also in the same collections package are potentially helpful odds and ends, including read-only dictionary and list collections, as well as read-only views of mutable dictionaries and lists. These are different from the read-only interfaces now supported in .NET, as they cannot be cast successfully to mutable dictionaries and lists. The CharHashSet class is an implementation of ISet for chars which generally outperforms HashSet<char>, and can be useful for certain specialized operations dealing with strings.

Some thoughts on global extension methods

New: Dynamically evaluate C# expressions and execute C# scripts with a single statement, from anywhere in a .NET application. Click here for more info.

In the past, I’ve sometimes found it useful to place some very generic extension methods in the global namespace, and/or at a visibility level corresponding to the classes extended. This can ease reuse, as the extension methods don’t require developers to take extra steps or have extra knowledge to use them; they’re just immediately available in type-ahead. Anyone who has suffered through using a namespace-happy API with multiple locations for extension methods may thank you for this approach.

As a brief example, consider the extension methods below. These mainly provide a bit of syntactic sugar for working with regular expressions, by converting what would often be multi-line calls into a single expression. The hiding of the System.Text.RegularExpressions namespace would be complete, except that in a few cases Match objects are exposed, as useful bundles of values and indices in the original string; System.Text.RegularExpressions would never need to be explicitly used even with those methods, if the var keyword were used.

using System.Collections.Generic;
using System.Text.RegularExpressions;

public static partial class StringExtensions
{
    private static MatchCollection EmptyMatchCollection = Regex.Matches("", "0");

    private static Match UnsuccessfulMatch = Regex.Match("a", "b");

    public static bool IsMatch(this string s, string pattern, bool isCompiled = false, bool ignoreCase = false)
    {
        if (s == null || pattern == null) return false;
        RegexOptions options = RegexOptions.CultureInvariant;
        if (ignoreCase)
            options |= RegexOptions.IgnoreCase;
        if (isCompiled)
            options |= RegexOptions.Compiled;
        return Regex.IsMatch(s, pattern, options);
    }

    public static int IndexOfPattern(this string s, string pattern, bool isCompiled = false, bool ignoreCase = false)
    {
        if (s == null || pattern == null) return -1;
        RegexOptions options = RegexOptions.CultureInvariant;
        if (ignoreCase)
            options |= RegexOptions.IgnoreCase;
        if (isCompiled)
            options |= RegexOptions.Compiled;
        Match match = Regex.Match(s, pattern, options);
        return match.Success ? match.Index : -1;
    }

    public static int LastIndexOfPattern(this string s, string pattern, bool isCompiled = false, bool ignoreCase = false)
    {
        if (s == null || pattern == null) return -1;
        RegexOptions options = RegexOptions.CultureInvariant | RegexOptions.RightToLeft;
        if (ignoreCase)
            options |= RegexOptions.IgnoreCase;
        if (isCompiled)
            options |= RegexOptions.Compiled;
        Match match = Regex.Match(s, pattern, options);
        return match.Success ? match.Index : -1;
    }

    public static MatchCollection Matches(this string s, string pattern, bool isCompiled = false, bool ignoreCase = false)
    {
        if (s == null || pattern == null) return EmptyMatchCollection;

        RegexOptions options = RegexOptions.CultureInvariant;
        if (ignoreCase)
            options |= RegexOptions.IgnoreCase;
        if (isCompiled)
            options |= RegexOptions.Compiled;

        return Regex.Matches(s, pattern, options);
    }

    public static IEnumerable FindAll(this string s, string pattern, bool isCompiled = false, bool ignoreCase = false)
    {
        if (s == null || pattern == null) return new List();

        RegexOptions options = RegexOptions.CultureInvariant;
        if (ignoreCase)
            options |= RegexOptions.IgnoreCase;
        if (isCompiled)
            options |= RegexOptions.Compiled;
        
        return Regex.Matches(s, pattern, options).Select(m => m.Value);
    }

    public static Match Match(this string s, string pattern, bool isCompiled = false, bool ignoreCase = false)
    {
        if (s == null || pattern == null) return UnsuccessfulMatch;

        RegexOptions options = RegexOptions.CultureInvariant;
        if (ignoreCase)
            options |= RegexOptions.IgnoreCase;
        if (isCompiled)
            options |= RegexOptions.Compiled;

        return Regex.Match(s, pattern, options);
    }

    public static Match MatchLast(this string s, string pattern, bool isCompiled = false, bool ignoreCase = false)
    {
        if (s == null || pattern == null) return UnsuccessfulMatch;

        RegexOptions options = RegexOptions.CultureInvariant | RegexOptions.RightToLeft;
        if (ignoreCase)
            options |= RegexOptions.IgnoreCase;
        if (isCompiled)
            options |= RegexOptions.Compiled;

        return Regex.Match(s, pattern, options);
    }

    public static string Find(this string s, string pattern, bool isCompiled = false, bool ignoreCase = false)
    {
        if (s == null || pattern == null) return "";

        RegexOptions options = RegexOptions.CultureInvariant;
        if (ignoreCase)
            options |= RegexOptions.IgnoreCase;
        if (isCompiled)
            options |= RegexOptions.Compiled;

        Match match = Regex.Match(s, pattern, options);
        return match.Success ? match.Value : "";
    }

    public static string FindLast(this string s, string pattern, bool isCompiled = false, bool ignoreCase = false)
    {
        if (s == null || pattern == null) return "";

        RegexOptions options = RegexOptions.CultureInvariant | RegexOptions.RightToLeft;
        if (ignoreCase)
            options |= RegexOptions.IgnoreCase;
        if (isCompiled)
            options |= RegexOptions.Compiled;
        Match match = Regex.Match(s, pattern, options);
        return match.Success ? match.Value : "";
    }

}

The MatchCollection class itself is kind of an odd duck, as it’s a useful collection that was never fully updated for use with Linq. If one makes frequent use of regular expressions it might be useful to also put a few simple Linq-enabling extension methods on MatchCollection itself; it’s up to the user whether to make them global. I leave it to the reader to implement the rest of the Linq methods as deemed useful in a particular case.

using System;
using System.Collections.Generic;
using System.Linq;
using System.Text.RegularExpressions;

public static partial class MatchCollectionExtensions
{
    public static bool Any(this MatchCollection source, Func<Match, bool> predicate)
    {
        return source.Cast<Match>().Any(predicate);
    }

    public static Match First(this MatchCollection source)
    {
        return (source == null || source.Count == 0) ? null : source[0];
    }

    public static Match First(this MatchCollection source, Func<Match, bool> predicate) {
        return source.Cast<Match>().First(predicate);
    }

    public static Match FirstOrDefault(this MatchCollection source)
    {
        return (source == null || source.Count == 0) ? null : source[0];
    }

    public static Match FirstOrDefault<T>(this MatchCollection source, Func<Match, bool> predicate)
    {
        return source.Cast<Match>().FirstOrDefault(predicate);
    }

    public static Match Last<T>(this MatchCollection source)
    {
        return (source == null || source.Count == 0) ? null : source[0];
    }

    public static Match Last(this MatchCollection source, Func<Match, bool> predicate)
    {
        return source.Cast<Match>().Last(predicate);
    }

    public static Match LastOrDefault(this MatchCollection source)
    {
        return (source == null || source.Count == 0) ? null : source[source.Count - 1];
    }

    public static Match LastOrDefault<T>(this MatchCollection source, Func<Match, bool> predicate)
    {
        return source.Cast<Match>().LastOrDefault(predicate);
    }

    public static IOrderedEnumerable<Match> OrderBy<Match, TKey>(this MatchCollection source, Func<Match, TKey> keySelector)
    {
        return source.Cast<Match>().OrderBy(keySelector);
    }

    public static IOrderedEnumerable<Match> OrderBy<Match, TKey>(this MatchCollection source, Func<Match, TKey> keySelector, IComparer<TKey> comparer)
    {
        return source.Cast<Match>().OrderBy(keySelector, comparer);
    }

    public static IOrderedEnumerable<Match> OrderByDescending<Match, TKey>(this MatchCollection source, Func<Match, TKey> keySelector)
    {
        return source.Cast<Match>().OrderByDescending(keySelector);
    }

    public static IOrderedEnumerable<Match> OrderByDescending<Match, TKey>(this MatchCollection source, Func<Match, TKey> keySelector, IComparer<TKey> comparer)
    {
        return source.Cast<Match>().OrderByDescending(keySelector, comparer);
    }

    public static IEnumerable<Match> Reverse(this MatchCollection source)
    {
        return source.Cast<Match>().Reverse();
    }

    public static IEnumerable<TResult> Select<TResult>(this MatchCollection source, Func<Match, TResult> selector)
    {
        return source.Cast<Match>().Select(selector);
    }

    public static IEnumerable<TResult> Select<TResult>(this MatchCollection source, Func<Match, int, TResult> selector)
    {
        return source.Cast<Match>().Select(selector);
    }

    public static IEnumerable<Match> Skip(this MatchCollection source, int count)
    {
        return source.Cast<Match>().Skip(count);
    }

    public static IEnumerable<Match> Take(this MatchCollection source, int count)
    {
        return source.Cast<Match>().Take(count);
    }

    public static Dictionary<TKey, Match> ToDictionary<Match, TKey>(this MatchCollection source, Func<Match, TKey> keySelector)
    {
        return source.Cast<Match>().ToDictionary(keySelector);
    }

    public static Dictionary<TKey, Match> ToDictionary<Match, TKey>(this MatchCollection source, Func<Match, TKey> keySelector, IEqualityComparer<TKey> comparer)
    {
        return source.Cast<Match>().ToDictionary(keySelector, comparer);
    }

    public static Dictionary<TKey, TElement> ToDictionary<Match, TKey, TElement>(this MatchCollection source, Func<Match, TKey> keySelector, Func<Match, TElement> elementSelector)
    {
        return source.Cast<Match>().ToDictionary(keySelector, elementSelector);
    }

    public static Dictionary<TKey, TElement> ToDictionary<Match, TKey, TElement>(this MatchCollection source, Func<Match, TKey> keySelector, Func<Match, TElement> elementSelector, IEqualityComparer<TKey> comparer)
    {
        return source.Cast<Match>().ToDictionary(keySelector, elementSelector, comparer);
    }

    public static List<Match> ToList<T>(this MatchCollection source)
    {
        return source.Cast<Match>().ToList();
    }

    public static IEnumerable<Match> Where<T>(this MatchCollection source, Func<Match, bool> predicate)
    {
        return source.Cast<Match>().Where(predicate);
    }
}

The MemoryCache class is another fundamentally useful .NET class that has a bit of a clumsy API, which can be remedied by providing facades to the clunkier methods. Very often one merely wants to set a cache duration when setting an object, and get an object of a specified type safely.

using System;
using System.Runtime.Caching;

public static partial class MemoryCacheExtensions
{
    public static void Set(this MemoryCache cache, string key, object value, long durationMilliseconds)
    {
        if (cache == null || key == null) throw new ArgumentNullException();
        else if (durationMilliseconds <= 0) return;
        else if (value == null)
            try { cache.Remove(key); } catch {}

        DateTimeOffset expiration = DateTime.Now.AddMilliseconds(durationMilliseconds);
        cache.Set(key, value, expiration);
    }

    public static T Get<T>(this MemoryCache cache, string key, T defaultValue = default(T))
    {
        object value = cache.Get(key);

        if (value == null || !(value is T))
            return defaultValue;
        else
            return (T)value;
    }
}

The common thread in these examples is that even in the .NET FCL (and CoreFX for that matter), API design is not always optimal. A primary use case for extension methods is where a developer desires to add useful functionality to an API, but without the ability to control the API itself, and this applies to .NET fundamentals as easily as anything else. When one is adding those methods to a core class, it may make sense to make the extension methods themselves as visible as the core classes so extended–and here the efficiency/usability gains of broadening visibility can be tainting in a sense, as above where the MatchCollection extensions are made global by virtue of the fact they may be returned from the result of another global extension method on the String class.

This may not be everyone’s cup of tea, and in fact I don’t tend to do this for seldom-used, specialty classes ever. The String class is obviously a different case from even MatchCollection and the like. And whenever creating extension methods, but increasing with their visibility, one should strongly consider placing them in partial, wisely named classes to avoid naming collisions in the future.

Global.asax, Keeping the Magic Alive

In my efforts to retrofit an old Sitecore Web Forms application for caching which is safe for use with postback, etc. in an elegant way, I needed to review the full set of “magic” methods available in the Global.asax application file, which ASP.NET wires up at runtime.

As a reminder, make sure that you’ve included a script runat=”server” tag enclosing your code–you may have to restore this if deleted from, or never added to, an empty file. Confusion abounds on the web as to whether Global.asax works with ASP.NET MVC (it does), primarily because of this missing script tag.

The application- and session-specific event methods are:

Application_Start
Application_End
Application_Error
Session_Start
Session_End

The request-specific events are:

Application_BeginRequest
Application_AuthenticateRequest
Application_AuthorizeRequest
Application_ResolveRequestCache
Application_AcquireRequestState
Application_PreRequestHandlerExecute
Application_PreSendRequestHeaders
Application_PreSendRequestContent
Application_PostRequestHandlerExecute
Application_ReleaseRequestState
Application_UpdateRequestCache
Application_EndRequest

Sources:
https://web.archive.org/web/20071223170129/http://articles.techrepublic.com.com/5100-10878_11-5771721.html
http://sandblogaspnet.blogspot.com/2008/03/methods-in-globalasax.html