Using /delayload to specify dependent DLL path

Let’s say you have an ObjectARX module with an implicit dependency on another DLL. Your installer puts both your ARX module and the dependent DLL into a private application folder. Windows must be able to resolve the implicit DLL dependency at load time, otherwise it will fail to load your ARX module. Alas, Windows will not automatically search for the dependent DLL in your application folder, even though your ARX module is located there. Therefore your ARX module won’t load because the dependent DLL cannot be resolved.

To address this problem, you may be tempted to add your application folder to the AutoCAD support path so that Windows can find your dependent DLL. This is a Very Bad Idea – please don’t ever do it! It imposes a completely unnecessary burden on end users, and it’s not scaleable because it could cause the maximum support path length to be exceeded. A better solution is to change the implicit dependency into an explicit dependency by using the linker’s delayload feature.

Making the change is easy. First, change your ARX project’s linker settings to delayload the dependent DLL:


Next, implement a delayload hook in one of your source files to explicitly load the dependent DLL from the same folder as your ARX module:

#include <delayimp.h>

#pragma comment(lib, "delayimp")

HMODULE MyExplicitLoadLibrary( LPCSTR pszModuleName )
  if( lstrcmpiA( pszModuleName, "MyDependent.dll" ) == 0 )
    CHAR szPath[MAX_PATH] = "";
    //_hdllInstance is the HMODULE of *this* module
    DWORD cchPath = GetModuleFileNameA( _hdllInstance, szPath, MAX_PATH );
    while( cchPath > 0 )
      switch( szPath[cchPath - 1] )
        case '\':
        case '/':
        case ':':
      break; //stop searching; found path separator
    lstrcpynA( szPath + cchPath, pszModuleName, MAX_PATH - cchPath );
    return LoadLibraryA( szPath ); //call with full path to dependent DLL
  return NULL;

FARPROC WINAPI MyDliNotifyHook( unsigned dliNotify, PDelayLoadInfo pdli )
  if( dliNotify == dliNotePreLoadLibrary )
    return (FARPROC)MyExplicitLoadLibrary( pdli->szDll );
  return NULL;
extern "C" PfnDliHook __pfnDliNotifyHook2 = MyDliNotifyHook;

When you use this technique, you do have to ensure that the delayload hook is added before any function in the dependent DLL is called. This is not a problem in most cases, but it could be a consideration if your ARX module initializes global objects whose constructors must call functions in the dependent DLL. This is just one more reason why you should avoid global variables.

QuirkyPolyline: exposing foolish programmers

In AutoCAD, the AcDbCurve class defines a general parametrized curve. On a parametrized curve, points in space are mapped to real numbers (parameters) by an arbitrary function F(p) for values of p from the curve start parameter to the curve end parameter. Defining a curve this way simplifies working with complex 3D curves because one can work in the curve’s one dimensional “parameter space” instead of the more complex three dimensional cartesian space.

The choice for how to parametrize a curve entity is up to the implementor. Parameters are designed to be opaque, so consumers of parametrized curves must make no assumptions about how a specific curve implements them. The only thing guaranteed about the parameters of a curve is that all points on the curve map to a unique parameter in a continuous range from the starting parameter to the ending parameter.

AutoCAD entities such as lines, circles, arcs, polylines, rays, and others are derived from AcDbCurve. These basic curve entities have very simple and straighforward parameter mappings that can be reverse engineered with very little effort, so lazy programmers sometimes make assumptions that are not guaranteed to always be true. Even smart programmers often get tripped up on parametrized curves because they use parameter space when they should be using distance space (that is, distance along the curve). The AcDbCurve interface provides functions to translate from parameter space back and forth to distance space, as well as back and forth to three dimensional cartesian space.

Wise programmers always let the curve itself perform all translations to or from distance space or cartesian space so that their code never relies on the actual value of a parameter. Foolish programmers take shortcuts by making assumptions about parameter values. For example, it happens that in all past versions of AutoCAD, the parameter of the midpoint of a polyline segment is exactly halfway between the segment starting parameter and the segment ending parameter, and the segment starting and ending parameters are always equal to the index number of the corresponding vertex. A foolish programmer (e.g. here and here) may use this knowledge to calculate the midpoint of the segment by guessing it’s parameter. The foolish programmer’s code will fail when a case arises where that guess is wrong. It is the purpose of QuirkyPolyline to provide such a case.

QuirkyPolyline defines a QPOLY command that prompts for a selection set of lightweight polylines in AutoCAD, then converts the selected polylines to “quirky polylines” that expose a completely different parameter scheme. Quirky polylines look and act perfectly normal in all respects. If you save quirky polylines, they will be stripped of their quirkiness. There are no long term effects of making polylines quirky; it’s just a temporary change that wears off on its own.

Making quirky polylines is useful for testing code that works with parametrized curves. If your code makes assumptions that it shouldn’t, quirky polylines will expose the error of your ways. Over the years, I have had arguments with people who insisted that their assumption had been validated by extensive testing. My counterargument is that someone, somewhere (possibly even Autodesk itself) already has or eventually will generate a derived entity where the assumption is not true. In addition to providing a proving ground, QuirkyPolyline is designed to demonstrate to the naysayers that unusual parametrization implementations can exist in the wild.

The download contains ARX modules for AutoCAD 2000 through AutoCAD 2015. To use QuirkyPolyline, just load the module that matches your version and architecture of AutoCAD. Create a lightweight polyline with the PLINE command, then use the QPOLY command to convert the normal polyline to a quirky polyline. You can use the LIST command to verify that it did become quirky, but you should notice no other detectable changes.

[Note: the download has since been updated and moved to the ARX freebies page]

Here is a useful little lisp function that defines a PP command for displaying the parameter value of a picked point on a selected curve entity. You can use this to explore other curve entities, and to see how the polyline’s parametrization changes after it becomes quirky.

(defun C:PP (/ sel pointoncurve param)
(if (setq sel (entsel))
(setq pointoncurve (apply 'vlax-curve-getclosestpointto sel))
(setq param (vlax-curve-getparamatpoint (car sel) pointoncurve))
(princ (strcat "nParam = " (rtos param)))

I hope this helps you avoid the pitfalls of parametric curve code. Remember, don’t be a foolish programmer!

RoboCache for Windows offline files cache management

I’ve written before about RoboCache, a command line utility I wrote for managing the offline files cache in Windows Vista, Windows 7, and now Windows 8. After using the utility internally for a few years, I decided that I might as well clean it up and make it available to the rest of you. RoboCache is now available at the ManuSoft web site. There is a shareware version available, and you can purchase the registered version at the ManuSoft store (cost is 25 USD for a single user license).

It’s nothing fancy, but it works great for my needs. The command syntax is modeled after the ROBOCOPY command, and is designed for handling an entire directory tree recursively, filtering files and folders by wildcard. My typical use case is running it from a batch file to pin remote Visual Studio project files on my laptop before travelling. Visual Studio project folders contain a lot of temporary build files and output files that don’t need to be (and therefore shouldn’t be) included in the cache. The goal is to pin only the necessary files, ignoring the ones that are not needed or will be recreated when the project is built. Here’s a sample call to recursively pin all files in the ‘Build’ folder (this is all one line in a batch file executed from the laptop):

ROBOCACHE Build /op:pin /s /xd debug* release* x64 Win32 .svn _* obj bin ipch /xf *.log *.tlog *._ls *.ncb *.user *.suo *.aps *.ilk *.pch BuildLog.htm *.err *.dmp *.pdb !*.bat #*.bat *.chm *.dia *.aps *.lnk *.Res.dll *.zip *log.txt *report.txt *.winmerge *.sdf *.opensdf

Programmers might notice that the /xd (ignore directories) and /xf (ignore files) wildcard lists look very similar to what one might encounter in a Subversion commit script. In fact, when I created the batch file I just copied and pasted from my SVN ignore lists. For me, the benefit of using RoboCache in this way is that only the minimum needed files are cached. After working on the cached files while I’m out of the office, my changes automatically sync to the folders on my desktop when I return to the office. I do use Subversion repositories for all my projects, and could just as well commit my remote changes to the repository, then update my desktop from the repository; however by using the offline files cache, the syncing all happens automatically and I never have to think about it.

RoboCache can do more than just pin and unpin files. It can perform any of ten operations on the target files (or any of four administrative functions on the files cache itself), including some operations that are not exposed through the Windows user interface. Below is a list of all currently supported operations (captured from the help screen, and not showing the available command line options for each operation).

/OP:cmd :: OPeration to perform (default is /OP:info).
:: info : display status info about the target(s)
:: pin : assure offline availability
:: unpin : unpin the target(s)
:: sync : synchronize cached files with remote files
:: rename : rename cached item (requires reboot)
:: delete : delete cached item
:: suspend : suspend the target folders (ignores files)
:: unsuspend : unsuspend the target folders (ignores files)
:: online : transition to online state
:: offline : transition to offline state
:: enable : enable offline files cache (ignores target)
:: disable : disable offline files cache (ignores target)
:: encrypt : encrypts offline files cache (ignores target)
:: decrypt : decrypts offline files cache (ignores target)

The installer simply adds RoboCache.exe to your system folder; no folders are created and no other changes are made. The shareware version of RoboCache displays a nag notification balloon in the system tray, but is otherwise completely functional. There is no separate documentation, however ROBOCACHE /? displays syntax and command options.

Polymorphic bit flags in C++

I often encounter cases in ObjectARX programming where numerous boolean flags need to be persisted as part of an AutoCAD database object. For filing and for passing around to other functions, it’s most efficient to package those bit flags into a single unsigned integer. Below is an example that demonstrates how to use a union inside a containing class that can be simultaneously used as either a single unsigned integer or as individual booleans.

// first declare a compact bitflag structure as bool values
struct FlagsAsBools
bool flag1 : 1;
bool flag2 : 1;
bool flag3 : 1;
bool flag4 : 1;
bool flag5 : 1;
bool flag6 : 1;
bool flag7 : 1;
bool flag8 : 1;
bool flag9 : 1;
bool flag10 : 1;
bool flag11 : 1;
bool flag12 : 1;

// now the containing class with useful constructors and operators
class Flags
// and the hidden union that actually contains the flags
union _Flags
FlagsAsBools asBools;
Adesk::UInt32 asUInt; //space for up to 32 flags
} mFlags;
mFlags.asUInt = 0;
Flags(Adesk::UInt32 flags)
mFlags.asUInt = flags;
Flags(const FlagsAsBools& flags)
mFlags.asBools = flags;
operator Adesk::UInt32 () const { return mFlags.asUInt; }
const FlagsAsBools& asBools() const { return mFlags.asBools; }
FlagsAsBools& asBools() { return mFlags.asBools; }

// now a class that contains the flags
class Settings
AcString msSomeStringSetting;
bool mbSomeBoolSetting;
Flags mFlags;
Settings () : mbSomeBoolSetting(false) {}

LPCTSTR someStringSetting() const { return msSomeStringSetting; }
void setSomeStringSetting(LPCTSTR someStringSetting) { msSomeStringSetting = someStringSetting; }
bool someBoolSetting() const { return mbSomeBoolSetting; }
void setSomeBoolSetting(bool someBoolSetting) { mbSomeBoolSetting = someBoolSetting; }
Flags flags() const { return mFlags; }
void setFlags(Flags flags) { mFlags = flags; }

// now use the flags as an argument to a function in different forms
void DoStuffBasedOnFlags( Adesk::UInt32 flags )
if( Flags(flags).asBools().flag7 )
{// do something when flag7 is set

void DoOtherStuffBasedOnFlags( const Flags& flags )
if( flags.asBools().flag1 )
{// do something when flag1 is set

void TestIt()
Settings& MySettings = GetSettings();
Flags flags = MySettings.flags();
if( flags.asBools().flag4 )
{// do something when flag4 is set
DoStuffBasedOnFlags( flags );
DoOtherStuffBasedOnFlags( flags );
flags.asBools().flag2 = true;
MySettings.setFlags( flags ); //save the new settings

Checking function return values

A lot of you are guilty of not checking return values from API function calls. I’ll bet you have a good excuse, like “it’s a lot of extra typing for no reason”, or “that function should never return an error code”.

I check return values religiously, use asserts liberally, and never assume that the implementation details of a function won’t change in the future. I think you should do the same.

In all my years of programming, I’ve never heard someone complain that their project failed because they spent too much extra time typing code to verify return values — but I have heard of projects failing because sloppy code made it almost impossible to diagnose and fix problems.

I see a lot of “sample code” posted on the internet in forums and blogs that omit error checking for brevity or to make the code clearer to read. That’s a bad idea, in my opinion. Sample code should be good code, not just clean code.

I will grant you that there are cases where error checking is not needed. For example, in the following code there is no need to verify the result of the call to acdbGetObjectId() because failure will be immediately evident in the call to acdbOpenObject() when idSelected is still null:

Acad::ErrorStatus MyFunction()
ads_name anameSelected;
ads_point apointSelected;
int nResult = acedEntSel( ACRX_T(“nSelect an object: “), anameSelected, apointSelected );
if( nResult != RTNORM ) return eNotAnEntity; //or whatever
AcDbObjectId idSelected;
acdbGetObjectId( idSelected, anameSelected ); //ignoring return value!
assert( !idSelected.isNull() ); //check it nevertheless in the debug build
AcDbEntity* pEnt = NULL;
Acad::ErrorStatus es = acdbOpenObject( pEnt, idSelected, AcDb::kForRead );
if( es != Acad::eOk ) return es;
// do something with pEnt
//I confess that I don’t bother checking the
//result of AcDbObject::close() unless it matters
return Acad::eOk;

There are legitimate cases where error checking is not needed, but I think especially sample code posted on the internet should encourage good programming habits by checking the result of every function that returns one.