Fun with multi-homing

We have a bunch of Xserves running Mac OS X server at work. Most of the machines are only using one of the 2 built in Ethernet ports as they are on the LAN. We have 2 machines that are customer facing and we just added a dedicated inbound connection for them. Being the clever person that I am, I decided that to ease the transition between the old and new IP blocks, I'd block the new connection into the second Ethernet port and we'd be good to go. Turns out it isn't that easy with the Darwin kernel. I setup the default connection to be the new network connection and traffic to the new IP addresses worked fine. However, traffic to the old address got hung up. After a lot of investigation, I determined it was due to Asymmetric routing. No problem, I thought, a few commands and it would work. I managed to do this in Linux by following an article, but it wasn't so easy in Mac OS X. Basically the traffic coming in to the old IP block had the responses going out through the other Ethernet interface out over the new IP block. Many routers block this as it kind of looks like an attack of sorts.

OK, so now that I was stuck, what would I do? I spent about 12 hours on this issue and through some magic use of Apache proxies and another server, I was able to get things working. It isn't pretty, but it solves the problem until we can get an A record changed (we have no control over that record).

Lesson learned, really, really, think carefully before deploying Mac OS X as a server; it can be quite frustrating to do things that I believe should be simple. Maybe I expect too much and OS X Server isn't designed for me.

Misinformed Author

As is pretty routine in our lousy local paper, I read an article where the author didn't bother to do his research. He lumps LinkedIn with MySpace and Facebook as social networking sites routinely used by those under 18. Here is my letter to the editor:

In your article today (Saturday, March 7) about social networking sites, your author has clearly failed to due his research as there is a glaring mistake in the first sentence. The author lumps LinkedIn with MySpace and Facebook. Anyone that has used LinkedIn or visited its home page knows that it is for professionals and is quite unlike Facebook and MySpace. In fact, I suspect that very few people under the age of 18 use the site. I've used LinkedIn for several years as it has helped me establish and maintain professional contacts. In fact, I routinely receive job inquiries and last summer was offered a job based on contacts through LinkedIn. LinkedIn is clearly not the same as MySpace and Facebook.

Accepting responsibility for bugs

When I make a mistake, I take full responsibility for the issue. While some people think I never make mistakes, I am human. The same goes for bugs in my software; software will never be perfect and I make my share of mistakes. I acknowledge these mistakes in my release notes where I say "Fixed". If it wasn't broken, then how could I fix it?

Today I had a user get on my case about me blaming others for bugs in ReceiptWallet. There are basically only 2 cases where I do this. The first is a bug where if "write metadata to PDFs" is turned on, sometimes PDFs will become corrupt. This one is quite easy to blame Apple. I've had users send me the original documents in question, I opened them in Preview (an application that Apple wrote), showed the Inspector, added some keywords, saved the document and then re-opened it. The result was that the document was corrupt. Recently I had a user report a very odd behavior in which the document didn't look corrupt, but when he tried to copy information out of it, the copied text was garbage. Sure enough, I did the above steps in Preview and demonstrated the problem. Sure sounds like an Apple bug to me; I've reported this to Apple a few times and have seen it fixed in the next OS release. Unfortunately it keeps coming back; what that means to me is that Apple's CoreGraphics routines for handling PDFs are not tolerant enough. Some random program implements PDF writing and then Apple has to handle it; so while I blame Apple, maybe I should blame the original PDF writing application. That, of course, would be fruitless as there are far too many of them out there. If Apple could work on their PDF handling, this issue might go ago. In any case, turning off "write metadata to PDFs" works around this bug. (I love this feature as it lets my wife enter receipts, email me the archive, and then I can just import them into ReceiptWallet with all the information intact.) This issue is clearly not an issue with ReceiptWallet except for the fact that ReceiptWallet has this option.

The second issue has to do with me blaming scanner vendors for their poor quality drivers. Anyone that has had a scanner on the Mac for any length of time knows how crappy the drivers are. I finally had my father acknowledge this after he fought with the HP drivers for the longest time (I wrote about this last August when the scanner driver reported that it used 16815 bits per pixel which means it had a ton of color information!). These drivers are typically ports from Mac OS 9; some of the UIs still look like OS 9. I see com.epson in crash log all the time; anyone that reads crash logs knows that this is in the bundle that ReceiptWallet has to load in order to run the scanner. In addition, I've lately seen crashes when ReceiptWallet quits.

Thread 0 Crashed:
0   libobjc.A.dylib               	0x905eb688 objc_msgSend + 24
1   com.apple.CoreFoundation      	0x93f9f372 CFBundleGetIdentifier + 50
2   com.apple.CoreFoundation      	0x93fa2dc5 __CFBundleDeallocate + 53
3   com.apple.CoreFoundation      	0x93fff768 _CFRelease + 216
4   ???                           	0x00fc5340 0 + 16536384
5   dyld                          	0x8fe13083 ImageLoaderMachO::doTermination(ImageLoader::LinkContext const

After some investigation and looking at Console logs for the users having this problem, there is clearly a message in it saying that the EPSON plugin is releasing its bundle identifier when it should not be and that the user should report this to the plugin vendor. Who should I blame in this case?

The only scanners I recommend are the Fujitsu ScanSnap and the Pentax DSMobile 600. The ScanSnap series don't use TWAIN drivers so they can't blow up ReceiptWallet and the Pentax DSMobile has incredibily well put together drivers. My guess is that they didn't start from legacy code. I've also had an engineer at a major scanner vendor acknowledge that their drivers are pretty poor, but unfortunately it will take time to fix. (Of course, I can't tell you which vendor.) I'd love for scanner vendors to either fix their drivers or acknowledge that they are poorly written; however, most vendors bundle some of their own software and barely test the TWAIN interface (Pentax doesn't bundle any Mac software, so they are totally dependent on TWAIN).

So while I'd love to be able to accept responsibility for these issues in ReceiptWallet and fix them, there is nothing I can do about it (I had one user suggest that I write scanner drivers for all the scanners...after I got up off the floor from rolling around laughing, I told the user that that was not going to happen).

ScreenSteps, cool idea, but not quite a Mac app

On Friday I stumbled across ScreenSteps, a program for doing documentation by quickly and easily capturing screen shots. While I no longer do documentation for my own software, I am starting to do some internal documentation for work on things like connecting to our VPN, setting up connecting to a file server, etc. I put together one "lesson" in maybe 15 minutes and was able to stick it on our internal web site. One thing that bothered me was that there were some dialogs that didn't look like Mac dialogs like the following (The icon looks funny and it should have been a sheet):

Things like this usually indicate it was written in REALbasic. While I don't have anything against REALbasic itself, I have problems with people slapping together programs in REALbasic and passing them off as Mac applications; in most cases they work like Mac applications, but they seem unpolished. I realize that this is a generalization, but I haven't seem a REALBasic based program that I'd purchase (at least knowingly). ScreenSteps didn't have the REALbasic libraries buried in its app bundle, so I knew it wasn't built using REALbasic. So did the developers purposely create crappy looking buttons? I don't think so. They appear to release the Mac and Windows versions at the same time, so they must be using some type of cross platform library. Ah, the about box gives me a clue. They're using a system called Runtime Revolution. Uggh, another attempt at creating cross platform applications with the click of a button.

So the question boils down to, can I live with the sort of Mac like appearance to quickly build documentation or can I find another tool? I'm not sure, but I'm definitely not a fan of cross platform tools as they produce apps that just aren't up to Mac UI guidelines.