Le 19/08/2016 à 11:43, Thomas Schmitt a écrit :
Pascal Hambourg wrote:Any _decent_ BIOS should ignore the partition table. It is none of its business.Yep. But there seems to be the tendency to install MBR code which extents the BIOS by looking at the partition table and hop onto the x86 code at the start of the partition marked by the "Boot" flag. So there was a risk that some well meant default mechanism decided to replace the ISOLINUX x86 code by such chain loader code.
As I previously wrote, such ill behaviour from a partition editor falls obviously in the "badly broken" category.
A Debian isohybrid MBR together with invalid GPT and a valid APM (Apple Partition Map) is quite far from looking sane. These features are fishing for boot firmware, not for the applause of partition editors. That's why they are so much borderline. A partition editor could come to the idea to make the GPT valid by patching in its standard Protective MBR template. It could come to the idea to correct the insane size value or the other fields of the APM Block0, which are actually x86 instructions to let the CPU hop on the ISOLINUX El Torito boot image.
You have a point, I didn't consider possible issues with the multiple invalid partition tables in the ISO hybrid image. I observed once that parted destroyed a hybrid MBR partition table I had created with gdisk (in order to boot Windows from BIOS on a GPT disk), and replaced it with a standard protective MBR.
Well, with fdisk "n" it seems to be ok, currently.
Yes, fdisk is definitely the tool of choice for this task. Parted just tries to be too clever.