llama: fix integer type consistency in split helpers

[MaheshJakkala]

This PR fixes integer type inconsistencies in llama_split_path
and llama_split_prefix.

Changes:

• Replace API-facing int with int32_t
• Avoid narrowing conversions from size_t
• Make signedness and intent explicit

Motivation:

• Aligns with llama : integer type consistency in llama.h #4574
• Prevents truncation and signed/unsigned mismatches
• Improves safety for downstream bindings

No behavior or API changes intended.

[taronaeo]

Why is your commit in the future?

[MaheshJakkala]

Thanks for pointing this out!
This was caused by an incorrect local system clock.
I’ve amended the commit to fix the timestamp and force-pushed the update.

[taronaeo]

Generally looks fine. I'll defer my review to @ggerganov as this affects libllama.

[ggerganov]

I had some minor style fixes somewhere, but I think the branch was write-protected and wasn't able to push.

diff

diff --git a/src/llama.cpp b/src/llama.cpp
index b3275ecae..eca063ecc 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -1116,7 +1116,6 @@ int32_t llama_split_path(
     return (int32_t) written;
 }
 
-
 int32_t llama_split_prefix(
     char * split_prefix,
     size_t maxlen,
@@ -1128,8 +1127,8 @@ int32_t llama_split_prefix(
 
     char postfix[32];
     snprintf(postfix, sizeof(postfix), "-%05d-of-%05d.gguf", split_no + 1, split_count);
-    const std::string str_postfix(postfix);
 
+    const std::string str_postfix(postfix);
     if (str_split_path.size() <= str_postfix.size()) {
         return 0;
     }
@@ -1139,13 +1138,13 @@ int32_t llama_split_prefix(
     if (str_split_path.compare(size_prefix, std::string::npos, str_postfix) == 0) {
         const size_t copy_len = std::min(size_prefix + 1, maxlen);
         snprintf(split_prefix, copy_len, "%s", split_path);
+
         return (int32_t) size_prefix;
     }
 
     return 0;
 }
 
-
 const char * llama_print_system_info(void) {
     static std::string s;
     s.clear(); // Clear the string, since it's static, otherwise it will accumulate data from previous calls.

[MaheshJakkala]

Thanks!
I’ve applied the minor style fixes you suggested and pushed the update.
Please let me know if anything else is needed.

[MaheshJakkala]

Thanks for the review!

I’ve fixed the EditorConfig issues (trailing whitespace) and pushed the update.

The remaining CI failures appear to be limited to server builds
(Linux & Windows) failing CMake configuration due to missing CURL
("Could NOT find CURL").

This PR does not touch server or CMake logic, so these failures seem
unrelated to the integer-type changes here.

Please let me know if you’d like me to adjust anything on my side,
or if these jobs can be retried / handled upstream.

[taronaeo]

The remaining CI failures appear to be limited to server builds
(Linux & Windows) failing CMake configuration due to missing CURL
("Could NOT find CURL").

Please try to rebase with master.

[MaheshJakkala]

I rebased the change cleanly on latest master and pushed a fresh branch
(fix-integer-types-split-clean) to avoid workflow / CI permission issues.
I’ll open a new PR from the clean branch.
Thanks for the review and guidance.