Why a whole chapter on error handling? The answer is that error handling with scripts that produces images requires a slightly different approach than normal error messages. In order to appreciate the added complexity one needs to consider how a graph script is called. Usually a graph script is called from an <img> tag. This means that the client (e.g. the browser) expect image data to be sent from the script. If we instead send a textual error message the browser will not be able to display anything. It was expecting a stream with image data and the data received was a representation of an error message which of course cannot be interpreted as an image.

To handle this problem the error messages produced by the library are normally generated as an image. A typical error message can be seen in Figure 6.1. Typical image error message

There is one exception to this rule and that is in the case of the "Headed already sent" error message. This means that the client has already received an header and with probability bordering on certainty this is a text header. Hence in this case it does not make sense to send an image so this, and only this, error message is sent as a normal "text/html" type message. A typical example how this message is shown in the browser can be seen in Figure 6.2. The "Header already sent error message"

When you get this error (and we mean when - not if) take a very close look at the text in red. This will tell you in what file and what line the erroneous output started. In most cases this will be an innocent looking space or a tab character. It could also be caused by multiple newlines at the end of a file. Since by definition these mistakes can be hard to spot since spaces and tabs are not normally visible in an editor so take care!