Embedded LOAD Error

[rgchris]

On attempting to load embedded Rebol data (Rebol 2 Manual on Scripts), the most basic case works just fine:

>> load/header "[Rebol [] 1 2]"
== [make object! [
    version: 0.0.0
    title: "Untitled"
    name: #(none)
    type: #(none)
    date: #(none)
    file: #(none)
    author: #(none)
    needs: #(none)
    options: #(none)
    checksum: #(none)
] 1 2]

However, if there's content beyond the end of the embedded data, LOAD returns an error:

>> load/header "[Rebol [] 1 2] 3"

** Script error: invalid /part count: #{2033}
** Where: copy unless case load
** Near: copy/part data end

It seems the culprit is this line here: src/mezz/sys-load.reb#L338. It's not clear what the intent here is:

• It appears to be trying to reassign data to #{2033} which would be incorrect
• It could be that the line was intended to read data: copy/part source end which would imply that data would be loaded again in the next case clause, however this can break:
  • if the source argument isn't binary, or
  • if the content has preamble:

    >> load/header "0^/[Rebol [] 1 2] 3"
    == [make object! [
        version: 0.0.0
        title: "Untitled"
        name: #(none)
        type: #(none)
        date: #(none)
        file: #(none)
        author: #(none)
        needs: #(none)
        options: #(none)
        checksum: #(none)
    ] 0
    [Rebol [] 1 2]]
    

    In this example, the header is correct, but the data is corrupted in multiple ways (0, extra block, Rebol [] as data)

---

In prior versions of Rebol, the preamble and appendix are discarded which is likely the correct outcome given the scope of load:

>> load/header "0^/[Rebol [] 1 2] 3"
== [make object! [
        title: "Untitled"
        name: none
        type: none
        version: none
        date: none
        file: none
        author: none
        needs: none
        options: none
        checksum: none
    ] 1 2]

---

Two cases that should resolve to true:

parse load/header "0^/[Rebol [] 1 2] 3" [
    object! integer! integer!
]

parse load/header #{92AF0A5B5265626F6C205B5D203120325D03} [
    object! integer! integer!
]

[Oldes]

Looks like a bug related to this one:
Oldes/Rebol-issues#1941

[rgchris]

Ah, I see—the intent is being able to sign scripts. That does muddy things a little. I wonder what portion of a script should be signed then? I'd assume:

"0^/[Rebol [] 1 2] 3"
-> ["0^/[" "Rebol [] " "1 2" "] 3"]
-> [preamble header signed-script appendix]

Assuming as the header stands apart from the script as it contains the signature...

It seems current behaviour is:

"0^/[Rebol [] 1 2] 3"
-> ["0^/[" "Rebol []" "0^/[Rebol [] 1 2] 3"]
-> [preamble header script]

Then there's also the null character termination, presumably:

"#!/thing^/Rebol [] 1 2 ^@ 3"
-> ["#!/thing^/" "Rebol [] " "1 2 " "^@" " 3"]
-> [preamble header script terminator appendix]

[Oldes]

All this module related code (load is one of its core parts) should be marked as "Hic Sunt Leones". I think that Brian never finished it. There is quite a lot of open related issues.

[Oldes]

My opinion is, that having a checksum in a script header is not secure enough... it may be used only to validate that the file was not modified by some transfer issue. Not that it is from a trusted source.

[acook]

Checksums/hashes are fine for ensuring data hasn't been accidentally corrupted. But an HMAC would be preferred as a way to ensure that data hasn't been intentionally modified. This requires that the user have access to the public key of the creator though.