refactor: store positions as a 64 bits integer in flatfiles #36669

pull darosior wants to merge 2 commits into bitcoin:master from darosior:2610_64bit_blockstorage changing 9 files +54 −23

darosior commented at 4:01 PM on October 26, 2026: member

That field is the last 32-bit in flat files management that is only bounded indirectly by callers' usage (unlike the file number for instance). In the spirit of the ongoing effort to avoid unnecessarily using 32 bits integers or platform specific types (for instance #24214, #33724, #34109, #35097, #35616), i was scanning for fields it could be prudent to promote to prevent any possible future misuse. This one is a good candidate because it does not change on-disk representation, since we serialize positions as varints.

flatfile: store position as a 64 bits integer

This is the last field that is only bounded by the usage callers make of it
(unlike the file number, for instance). Promoting it to a 64 bits integer also
does not change on-disk representation since we serialize positions as varints.
Therefore, widen them to prevent any possible future misuse.

17556cb1f5

flatfile: use 64-bit file positioning syscalls in Windows 48092a9042
DrahtBot added the label Refactoring on Oct 26, 2026
DrahtBot commented at 4:01 PM on October 26, 2026: contributor


The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.



External sites
- Code coverage & benchmarks: https://corecheck.dev/bitcoin/bitcoin/pulls/36669
- Ralph review <img src=https://review.fish.foo/assets/bitcoin/bitcoin/36669.png alt="Ralph review status"> https://git.fish.foo/bitcoin/bitcoin/pulls/36669#author-ralph


Reviews

See the guideline and AI policy for information on the review process.

Type Reviewers

ACK sedited, hebasto, instagibbs, l0rinc

If your review is incorrectly listed, please copy-paste <code></code> into the comment that the bot should ignore.
sedited approved
sedited commented at 4:28 PM on October 26, 2026: contributor

tACK 48092a9042d8fcfd6bb7ae5911a1d53b8746d546

Also ran a few extra tests locally to stresss this a bit and tested and reproduced the windows code.
hebasto approved
hebasto commented at 4:34 PM on October 26, 2026: member

ACK 48092a9042d8fcfd6bb7ae5911a1d53b8746d546, I have reviewed the code and it looks OK.
instagibbs approved
instagibbs commented at 4:36 PM on October 26, 2026: member

ACK 48092a9042d8fcfd6bb7ae5911a1d53b8746d546

reproduced on linux only

Type	Reviewers
ACK	sedited, hebasto, instagibbs, l0rinc

in src/chain.h:93 in 48092a9042

l0rinc commented at 4:51 PM on October 26, 2026:

48092a9 flatfile: use 64-bit file positioning syscalls in Windows:

This increases sizeof(CBlockIndex) slightly, we can bring that back in a followup PR by moving nTx after m_chain_tx_count.

<details> <summary>Restore the CBlockIndex size</summary>

diff --git a/src/chain.h b/src/chain.h
index 830cabe3f6..b2380468b2 100644
--- a/src/chain.h
+++ b/src/chain.h
@@ -117,17 +117,17 @@ public:
     //! (memory only) Total amount of work (expected number of hashes) in the chain up to and including this block
     arith_uint256 nChainWork{};
 
-    //! Number of transactions in this block. This will be nonzero if the block
-    //! reached the VALID_TRANSACTIONS level, and zero otherwise.
-    //! Note: in a potential headers-first mode, this number cannot be relied upon
-    unsigned int nTx{0};
-
     //! (memory only) Number of transactions in the chain up to and including this block.
     //! This value will be non-zero if this block and all previous blocks back
     //! to the genesis block or an assumeutxo snapshot block have reached the
     //! VALID_TRANSACTIONS level.
     uint64_t m_chain_tx_count{0};
 
+    //! Number of transactions in this block. This will be nonzero if the block
+    //! reached the VALID_TRANSACTIONS level, and zero otherwise.
+    //! Note: in a potential headers-first mode, this number cannot be relied upon
+    unsigned int nTx{0};
+
     //! Verification status of this block. See enum BlockStatus
     //!
     //! Note: this value is modified to show BLOCK_OPT_WITNESS during UTXO snapshot

</details>

l0rinc approved
l0rinc commented at 4:52 PM on October 26, 2026: contributor

tested ACK 48092a9042d8fcfd6bb7ae5911a1d53b8746d546

About time we got rid of the 32-bit file offsets - checked the CBlockIndex layout with Clang and GCC, the padding nit can be addressed in a follow-up.
fanquake merged this on Oct 26, 2026
fanquake closed this on Oct 26, 2026
fanquake commented at 5:04 PM on October 26, 2026: member

Backported to 32.x in #36658.
fanquake referenced this in commit 98f204f51b on Oct 26, 2026
fanquake referenced this in commit 09a3519e3b on Oct 26, 2026
fanquake commented at 5:19 PM on October 26, 2026: member

Backported to 31.x in #36661.
fanquake referenced this in commit d47e0d7e24 on Oct 26, 2026
fanquake referenced this in commit 307fbeeaf4 on Oct 26, 2026
sedited referenced this in commit f0de87f040 on Oct 26, 2026
fanquake referenced this in commit adb363c753 on Oct 26, 2026
fanquake referenced this in commit a9e7b35d06 on Oct 26, 2026
fanquake commented at 6:25 PM on October 26, 2026: member

Backported to 30.x in #36664.
fanquake referenced this in commit 241537a697 on Oct 26, 2026
fanquake commented at 7:42 PM on October 26, 2026: member

Backported to 29.x in #36666.
fanquake referenced this in commit 5c99cb6745 on Oct 26, 2026

Contributors

Labels

Linked (view graph)

#24214 Fix unsigned integer overflows in interpreter #33724 refactor: Return uint64_t from GetSerializeSize #34109 refactor: Use uint64_t over size_t for serialize corruption check in fees.dat #35097 util: Return uint64_t from _MiB and _GiB operators #35616 refactor: Use u64 over size_t for all cache sizes to fix a 32-bit overflow #36658 [32.x] Finalise 32.1 #36661 [31.x] Finalise 31.2 #36664 [30.x] Finalise 30.4 #36666 [29.x] Finalise 29.5

refactor: store positions as a 64 bits integer in flatfiles #36669

External sites

Reviews