info: Fix off-by-one in R_InitSpriteDefs

[pimzero]

According to its documentation, R_InitSpriteDefs expects a NULL terminated list of sprites names (namelist). By looking at the code it indeed get the length fo this list by looking for a NULL pointer.

By looking at the callgraph, we can see that namelist is a pointer to sprnames. However, sprnames is not a NULL-terminated list. Until today we were lucky to have a 0 soon after sprnames (in my case, I had the value 1, followed by a 0, so it was accessing invalid data)

Let's add some room in sprnames for this NULL pointer.

Found with https://github.com/lse/ref-k/pull/87

[xdbob]

nice but how tf, the OG linux-doom was not segfaulting ?

[pimzero]

My understanding is the following:

in linux doom:

• srpnames is not null-terminated in linux doom
• the next symbol in the data section is states, because it is just after sprnames in the source code
• the first entry of states is 8 bytes of 0, which, by accident or evil smartness, terminates the sprnames array

In kdoom: we build with --fdata-sections, the linker is well within its right to move things around - anything could have been deref. IMO we were just lucky to not hit this before.

I didn't manage to run linuxdoom so I can't try to replicate this behaviour on linux