SKP ZIP archive (SketchUp 2021 and later)

From byte 69 (0x45) to the end, the file is a ZIP archive as PKWARE’s APPNOTE describes it: the members one after another, each with a local header, then the central directory and the end record. SketchUp writes it with a few particulars, all seen in every file examined, set out below. All numbers are little-endian.

Layout

Part	Size	Contents
Local header	30 + name + 20	One before each member’s data, with a ZIP64 extra field.
Data	compressed size	The member, compressed with Deflate.
…		The next member, and so on; model.dat is always last.
Central directory	46 + name each	One record a member, in the same order.
End record	22	The member count and where the central directory is.

Offsets

Every offset in the archive, the local header offset in each central directory record and the directory offset in the end record, counts from the start of the ZIP archive, byte 0x45 of the file, not from the start of the file. To find a member’s local header, add 0x45 (or wherever the archive starts) to its offset.

Tools such as unzip notice the difference, warn of “69 extra bytes at beginning” and cope. The explorer looks for the central directory at the offset counted from the archive’s start, then from the file’s, so it also reads a file that another program has rewritten with offsets from the file’s start.

Local header

Offset	Size	Type	Name	Seen in SketchUp files
0	4	u32	Signature	`50 4B 03 04`, “PK” 3 4.
4	2	u16	Version needed	45 (4.5, the ZIP64 version).
6	2	u16	Flags	0 for the two PNGs in `meta/`; 6 for every other member. Bits 1 and 2 together mean Deflate’s “super fast” option.
8	2	u16	Method	8, Deflate, for every member.
10	2	u16	Time	0.
12	2	u16	Date	`0x0020`: in DOS form, year 1980, month 1, day 0. No real date is stored.
14	4	u32	CRC-32	The CRC-32 of the uncompressed member, filled in.
18	4	u32	Compressed size	The size of the data, filled in.
22	4	u32	Size	The size of the member once inflated, filled in.
26	2	u16	Name length	The length of the name.
28	2	u16	Extra length	20.
30	name length	text	Name	The member’s path, such as `meta/meta.dat`.
30 + name	20	bytes	Extra field	A ZIP64 extra field.

The data follows straight after the extra field. The sizes are in the header, so there is no data descriptor after it.

ZIP64 extra field

Every local header carries the same 20-byte extra field:

Offset	Size	Type	Name	Seen
0	2	u16	ID	1, the ZIP64 field.
2	2	u16	Length	16.
4	8	u64	Size	0.
12	8	u64	Compressed size	0.

01 00 10 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00

The ZIP64 extra field, with both 64-bit sizes zero

The real sizes are in the 32-bit fields of the header. A reader should use the 64-bit values only when a 32-bit field holds FFFFFFFF, as the ZIP specification says; the explorer does so.

Central directory

One record for each member, in the same order as the members:

Offset	Size	Type	Name	Seen in SketchUp files
0	4	u32	Signature	`50 4B 01 02`.
4	2	u16	Version made by	0.
6	2	u16	Version needed	20 (2.0).
8	2	u16	Flags	General purpose flags.
10	2	u16	Method	The compression method.
12	2	u16	Time	The DOS time.
14	2	u16	Date	The DOS date.
16	4	u32	CRC-32	The CRC-32 of the uncompressed member.
20	4	u32	Compressed size	The size of the data.
24	4	u32	Size	The size once inflated.
28	2	u16	Name length	The length of the name.
30	2	u16	Extra length	0: no extra field.
32	2	u16	Comment length	0: no comment.
34	2	u16	Disk	The disk the member starts on.
36	2	u16	Internal attributes	0.
38	4	u32	External attributes	0.
42	4	u32	Local header offset	Where the member’s local header is, counted from the start of the archive (see Offsets).
46	name length	text	Name	The member’s path.

When a size or the offset is FFFFFFFF, the real value is in a ZIP64 extra field in the record. None was seen; the explorer reads one if it is there.

End record

The last 22 bytes of the file: the end of central directory record, with no comment.

Offset	Size	Type	Name	Meaning
0	4	u32	Signature	`50 4B 05 06`.
4	2	u16	Disk	The number of this disk: 0.
6	2	u16	Directory disk	The disk the central directory starts on: 0.
8	2	u16	Entries on this disk	The number of members.
10	2	u16	Entries	The number of members.
12	4	u32	Directory size	The size of the central directory in bytes.
16	4	u32	Directory offset	Where the central directory starts, counted from the start of the archive.
20	2	u16	Comment length	0.

50 4B 05 06 00 00 00 00 41 00 41 00 12 14 00 00 E6 72 3B 00 00 00

The end record of a SketchUp 2023 file: 65 members, a 5,138-byte directory at 0x3B72E6 from the archive’s start, which is 0x3B732B in the file

To find it, search backwards from the end of the file for 50 4B 05 06; with no comment it is exactly 22 bytes from the end.

ZIP64 end records

An archive with more than 65,535 members, or a directory beyond 4 GB, needs the ZIP64 end of central directory record and its locator just before the end record. No SketchUp file seen has them, not even a 693 MB model with 2,720 members. The explorer reads them if they are there: the locator (50 4B 06 07, a u32 disk, a u64 offset of the ZIP64 end record, a u32 disk count) sits 20 bytes before the end record, and the ZIP64 end record (50 4B 06 06) gives 64-bit counts, size and offset of the directory.

Deflate

Every member is compressed with Deflate (method 8, RFC 1951). The explorer has its own decoder for it, which follows the steps below.

The PNG members: stored blocks

The two PNG pictures in meta/ are “compressed” as Deflate stored blocks: the PNG sits in the file as it is, behind a 5-byte block header. So the compressed size is the size plus 5.

01 AA 7C 55 83 89 50 4E 47 …

A stored block: the final-block bit with type 0, the length 31,914 (0x7CAA), its complement (0x8355), then the PNG

A stored block holds at most 65,535 bytes, so a bigger PNG needs several. SketchUp splits them smaller than that: one preview picture of 48,967 bytes was stored in 48,977 bytes, two blocks with their 5-byte headers. The explorer shows a PNG that is a single stored block straight from the file, and inflates and decodes the others.

Reading Deflate data

Deflate data is a sequence of blocks, read as a stream of bits, lowest bit of each byte first. Numbers in the stream are read lowest bit first too; Huffman codes are read one bit at a time from their first (highest) bit.

Read 1 bit: 1 means this is the last block.
Read 2 bits: the block type. 0 is stored, 1 fixed Huffman codes, 2 dynamic Huffman codes; 3 is an error.
Stored: skip to the next byte boundary, read LEN (u16) and NLEN (u16), check that NLEN is the complement of LEN, then copy LEN bytes to the output.
Fixed: use the fixed code lengths: 8 bits for literals 0 to 143, 9 for 144 to 255, 7 for 256 to 279, 8 for 280 to 287, and 5 bits for each of the 30 distance codes.
Dynamic: read HLIT (5 bits, plus 257 literal/length codes), HDIST (5 bits, plus 1 distance codes) and HCLEN (4 bits, plus 4). Read HCLEN 3-bit lengths for the code-length alphabet, in the order 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15. Build that code, and with it read HLIT + HDIST lengths: symbols 0 to 15 are a length; 16 repeats the previous length 3 to 6 times (2 extra bits); 17 gives 3 to 10 zeros (3 extra bits); 18 gives 11 to 138 zeros (7 extra bits). The first HLIT lengths are the literal/length code, the rest the distance code.
Build each Huffman code from its lengths: count the codes of each length, give the shortest codes the lowest values, and number the symbols of one length in order.
Decode symbols until 256. A symbol below 256 is a literal byte. 257 to 285 is a length (table below); read a distance code and its extra bits, then copy that many bytes from that far back in the output.
Repeat from step 1 until the last block is done.

Length codes	Extra bits	Lengths
257–264	0	3–10
265–268	1	11–18 (bases 11, 13, 15, 17)
269–272	2	19–34 (bases 19, 23, 27, 31)
273–276	3	35–66 (bases 35, 43, 51, 59)
277–280	4	67–130 (bases 67, 83, 99, 115)
281–284	5	131–257 (bases 131, 163, 195, 227)
285	0	258

Distance codes	Extra bits	Bases
0–3	0	1, 2, 3, 4
4–5	1	5, 7
6–7	2	9, 13
8–9	3	17, 25
10–11	4	33, 49
12–13	5	65, 97
14–15	6	129, 193
16–17	7	257, 385
18–19	8	513, 769
20–21	9	1025, 1537
22–23	10	2049, 3073
24–25	11	4097, 6145
26–27	12	8193, 12289
28–29	13	16385, 24577

A length or distance is its base plus the value of its extra bits. The members are large: model.dat was 17.5 MB to 86.6 MB once inflated in the files seen, so the explorer inflates at most the first 64 MB of it for its outline.

.SKP File Format – ZIP archive