The Bloatware Debate

Date: Fri, 30 April 1999
Forum: Risks-Forum Digest
Subject: The Bloatware Debate

One of the chief hallmarks of early UNIX was how simple, compact programs worked well together. Brian W. Kernighan's definition of a good program was a program so good and so consistent that it could be used for an entirely different purpose and be expected to work well. UNIX, they said, was a way of thinking more than an operating system. And, with Brian's Software Tools series, it was surely so.

Microsoft Windows is also a way of thinking - or not thinking, to be more exact. In almost every possible sense it is the anathema of the programming community, if that community still abides by and adheres to the solid thinking delineated by Brian so many years ago.

MS Windows programming is considered too difficult to attempt head on. Where we come from most major corporations, financial institutions and the like promised a smooth transition from UNIX or DOS to Windows 3.1x within a matter of weeks. Management talking of course. When they found this would not work they decided to invest heavily in 16-bit Visual Basic applications. Operative word "heavily". These bloatware masters sunk almost any machine made. Clearly this was not the answer either.

People looked to Kahn. Borland, with its Turbo C, saw the opening and released Borland C, and shortly thereafter Scott Randell who a year earlier had toured with MSC 7.0 (which admittedly never worked) was out rocking again, this time with Visual C++. The environment was unbelievable; the executables were extremely bloated; but still and all it was Microsoft talking, and still and all they were smaller than the corresponding Borland images. COBOL programmers everywhere were suddenly encouraged to learn C++, develop code browsing skills, learn about preprocessors, assembly language, CodeView and subsequent debuggers, and the world entered into a tailspin.

What originally started as a rather feeble but lucky attempt to get on the OO bandwagon, the MFC soon became something you'd like to see Steve McQueen kill. Patches and work-arounds and bugs and more bugs, and bloat and more bloat. The current splash screen module is a case in point: Microsoft includes a 16-color bitmap which weighs in at nearly 200KB for you. This bitmap can be compressed with RLE encoding to less than half that size. The idea of banging a 100KB splash bitmap in an application is still, however, sickening. Yet Microsoft gladly gives you the bitmap at 200KB, happy if you don't understand what you are doing by using it. Your application will be more sluggish than their own bloatware, and people will be less inclined to complain about what they themselves do.

Microsoft's RegClean, a popular product for fixing corruptions in the MS Windows Registry is another case in point. When this application was originally introduced I downloaded it and wondered about its size. It weighed in then at nearly a megabyte. Similar applications out there were 20KB and hardly more. What was inside this monster? I opened it and looked inside.

Remember all those stories about how surgeons in the old days just threw their rubber gloves inside the patient's stomach before sewing them back up again? Well here you had it. There were humungoid bitmaps never used. There were dozens of icons never referenced. There were tens of kilobytes of entries in the string table that had no meaning for the application whatsoever.

I honed the app down and came to the conclusion that the actual size of RegClean should be about 45KB. That as compared to its distribution size of nearly one megabyte. Clearly bloat is not only a question of adding features almost no one wants. Bloat is a condition of the mind, permeating software houses everywhere.

Clearly again the distribution of RegClean was highly irresponsible. But remember, MS Windows is not just an operating system - it is a way of thinking, or not thinking as you may have it. And it has permeated the entire industry today. Our hats off to Microsoft.

In conclusion: there are few application domains even today that require executables of over 100KB, and most ordinary tasks can be adequately managed by executables in the 20KB range. This is simply a fact.

There are no excuses. Either we think or we don't. There is no in between.

RA Downes Radsoft Laboratories ( http://www.radsoft.net/ )

Date: Sun, 02 May 1999 16:12:13 +0000
Forum: Risks-Forum Digest
Subject: Re: Bloatware Debate (Downes, RISKS-20.35)

A certain "Johnny" has written to me from Microsoft because of my posting in RISKS-20.35 about MS bloat. The tone was a thinly disguised threat. In his opening, "Johnny" stated that the "bloat" of MS RegClean was due no doubt to having static links. Discussing the sweeping ramifications of such a statement is unnecessary here. The mind boggles, it is sufficient to state. The MSVC runtime is a mere 250,000 bytes and in fact is not statically linked anyway to MS RegClean, AFAIK [as far as I know]. MS RegClean is an MFC app and will by default use the dynamically linked MFC libraries. And even if its static code links were an overhead here they would add but a small fraction of the total bloat, say 40KB at most.

For whatever reason, I decided to download the latest version of MS RegClean from BHS again and pluck it apart. This is what I found. I have tried - and it has been difficult - to keep subjective comments out of this report.

Current Status of RegClean Version 4.1a Build 7364.1

Image Size (Unzipped and ready to run): 837,632 bytes (818KB)

Import Tables

The import section in the PE header. This gives an indication of just how (in)effective the use of Bjarne's C++ has been. In this case, the verdict is: "pretty horrible". A walloping 7,680 bytes are used for the names of the relocatable Win32 imports. These are the actual names of the functions (supposedly) called. MS RegClean does not call most of these functions - they remain because an MFC template was originally used, most likely borrowed from another application, and it was never "cleaned". This is corroborated by what is found among the "Windows resources": over half a dozen standard menus, assorted graphic images, print preview resources, etc. that have nothing to do with the application at hand.

Resources

Please understand that resources not only bloat an executable with their own size, but with additional reference data, in other words the bloat factor of an unused or bad resource is always somewhat larger than the size of the bloating resource itself.

Accelerators

Sixteen (16) unused accelerators from an MFC template were found: Copy, New, Open, Print, Save, Paste, "Old Undo", "Old Cut", Help, Context Help, "Old Copy", "Old Insert", Cut, Undo, Page Up, Page Down. MS RegClean uses only one accelerator itself, not listed here.

Bitmaps

This was a particularly sorry lot. The main bloat here was a splash screen bitmap weighing in (no RLE compression of course) at over 150KB. Further, Ctl32 static library bitmaps were found, meaning MS RegClean is still linking with the old Ctl32v2 static library which was obsolete five years ago and which automatically adds another 41KB to the image size.

Cursors

Six (6) cursors were found, none of which have anything to do with this application.

Dialogs

A very messy chapter indeed. MS RegClean walks around with eighteen (18) hidden dialogs, of which only one or at the most two are ever used. The others are just - you took the words out of my mouth - junk. The findings (read it and weep):
  1. Eleven (11) empty dialogs with the caption "My Page" and the static text "Todo", all identical, all empty, and of course all unused. This is a wonder in and of itself.
  2. The main "wizard" dialog actually used by the application is left with comment fields to help the programmers reference the right controls in their code (subjective comment removed).
  3. A "RegClean Options" dialog which AFAIK is never used.
  4. A "New (Resource)" dialog, probably a part of the development process, just stuffed in the stomach at sew-up time and left there for posterity.
  5. A "Printing in Progress" dialog.
  6. A "Print Preview" control bar dialog.

Icons

MS RegClean has three icons, all with images of 48x48 in 256 colors (of course). The funniest thing here is that the authors of MS RegClean have extracted the default desktop icon from shell32.dll, which is available at runtime as a resident resource anyway and at no image bloat overhead at all, and included it in toto in their executable.

Menus

MS RegClean has eight (8) menus, at least half of these are simply junk left around by the MFC template. Another menu indicates that the authors of RegClean have in fact worked from an internal Microsoft Registry tool - rather bloated in itself it seems.

String Table(s)

Actually it need only be one string table, but Microsoft itself has never learned this. The findings here were atrocious. And you must remember that strings stored in a string table are stored in Unicode, which means that their bloat automatically doubles. Further, MS's way of indexing strings in a string table means a 512 byte header block must be created for every string grouping, and strings are grouped according to the high 12 bits of their numerical identifiers (yes they are 16-bit WORD identifiers). Meaning indiscriminate or random numbering of string table entries will make an otherwise innocent application literally explode.
347 (three hundred forty seven, yep, your video driver is not playing tricks on you) string table entries were found in MS RegClean, including 16 identical string entries with the MS classic "Open this document" as well as archaic MFC template toggle keys texts which are not used here (or almost anywhere else today). Most of these strings have - of course - nothing to do with the application at hand.

Toolbars

Toolbars are a funny MS way of looking at glyph bitmaps for use in toolbar controls. MS RegClean has two - one which may be used by the application, and one which was part of the original MFC template and never removed.

Total Accountable Resource Bloat

The total accountable (i.e. what can be directly calculated at this stage) resource bloat of MS RegClean 4.1a Build 7364.1 is over 360,000 bytes (350KB).

Total Accountable Code Bloat

Harder to estimate, but considering that most of the code is never used, only part of an MFC template that the authors of MS RegClean lack the wherewithal to remove, the original estimate of a total necessary image size of 45KB for the entire application must still stand.

In Conclusion

Bloat is not a technical issue, but verily a way of thinking, a "state of mind". Its cure is a simple refusal to accept, and a well directed, resounding "clean up your act and clean up your code!"

PS. Send feedback on RegClean to

RA Downes, Radsoft Laboratories http://www.radsoft.net