Some others compilers can be more advanced in the code generation for a new target. If you really need one of them, just say it to the GNU make with the CC or CXX variable and you are done.

You question tend to imply that the compiler and the IDE have some kind of dependency. While some commercial entity like to emphasis this claim, it’s rarely the case in practice. Most compilers can be used by any general IDE. Using the GNU make, it’s very easy to compile the same project for many targets using different compilers.

If you are using non UNIX machine for your dally task, you may find my point of view a bit strange. But if you routinely use a Linux system for example, you will find the GNU tools so natural that you don’t ever loose time to install an IDE.

Of course as with any risk mitigation evaluation it strongly depends on your project. If you are doing things that fall outside the standard use case for the tool your risk goes up. So if I were developing for a brand new chip, lets use the MSP430 when it first arrived, and had a customer facing project I would not suggest using the GNU tools. They typically show up late to the game because it is unusual for the manufacturer to assign engineers to implement support in free tools before the chip reaches the market. On the other hand I would believe the risk significantly lower to purchase the IAR compiler for the chip the day it reaches the market because they had the resources and access to the design early enough to add support and test it.

I use GNU tools frequently for embedded projects. On the other hand if I am doing a project that has very tight deadlines or is every expensive for the customer I will chose compilers for IAR or Kiel over GNU because of the reduced perceived risk.

On the other hand if you have a GNU expert on hand it can be very handy to be able to fix your compiler when you discover and issue. I have been on projects where I had to drop a commercial compiler for the GNU specifically because I could not get the fix and meet my deadline.

As far as GNU output it varies greatly based on platform and language. For example I have found that code I write for the Cortex-A8 in C and C++ the newest GNU compiler is actually a little better than RVCT 4.x. On the MSP430 I save significant space using the IAR over the GNU toolchain. It should also be mentioned that the GNU toolchain is a fast moving target. The code generated for the MSP430 between the last 3 versions has already reduced code size by 20% on my average programs and at this rate will be on par with IAR in another year or so.

This means that a company making (for example) compilers for low-end processors is likely to get a low unit price and a relatively small customer base. In these conditions it will be a real struggle to provide any tool, let alone a quality tool.

Compare this with the GNU tools – large developer base, large user base. Of course, Jon Titus is right – GNU tools are a good basis for many companies to develop their products on – AdaCore being one good example of this.