Make --log-disable work a bit better

llama.cpp/common.h

@@ -31,7 +31,7 @@
 #define die_fmt(fmt, ...) do { fprintf(stderr, "error: " fmt "\n", __VA_ARGS__); exit(1); } while (0)
 
 #define print_build_info() do {  \
-    fprintf(stderr, "%s: llamafile version " LLAMAFILE_VERSION_STRING "\n", __func__, LLAMA_BUILD_NUMBER, LLAMA_COMMIT);  \
+    LOG("%s: llamafile version " LLAMAFILE_VERSION_STRING "\n", __func__, LLAMA_BUILD_NUMBER, LLAMA_COMMIT);  \
 } while(0)
 
 // build info

llama.cpp/log.h

@@ -12,6 +12,8 @@
 #include <algorithm>
 #include <cinttypes>
 
+#include "llamafile/log.h"
+
 // --------------------------------
 //
 // Basic usage:
@@ -430,6 +432,7 @@ inline FILE *log_handler2_impl(bool change = false, LogTriState append = LogTriS
 // INTERNAL, DO NOT USE
 inline FILE *log_disable_impl()
 {
+    FLAG_log_disable = true;
     return log_handler1_impl(true, LogTriStateSame, LogTriStateTrue);
 }
 
@@ -439,6 +442,7 @@ inline FILE *log_disable_impl()
 // INTERNAL, DO NOT USE
 inline FILE *log_enable_impl()
 {
+    FLAG_log_disable = false;
     return log_handler1_impl(true, LogTriStateSame, LogTriStateFalse);
 }
 

llama.cpp/server/server.cpp

@@ -12,6 +12,7 @@
 #include "llama.cpp/ggml-cuda.h"
 #include "llamafile/llamafile.h"
 #include "llamafile/version.h"
+#include "llamafile/log.h"
 
 #define CPPHTTPLIB_NO_EXCEPTIONS 1
 #define CPPHTTPLIB_FORM_URL_ENCODED_PAYLOAD_MAX_LENGTH 1048576
@@ -2313,6 +2314,7 @@ static void server_params_parse(int argc, char **argv, server_params &sparams,
         }
         else if (arg == "--log-disable")
         {
+            FLAG_log_disable = true;
             log_set_target(stdout);
             LOG_INFO("logging to file is disabled.", {});
         }
@@ -3086,7 +3088,7 @@ int server_cli(int argc, char ** argv) {
         }
     }
 
-    tinyprint(2, "loading weights...\n", NULL);
+    tinylog("loading weights...\n", NULL);
 
     LOG_INFO("HTTP server listening", {
                                           {"hostname", sparams.hostname},

llamafile/cuda.c

@@ -31,6 +31,7 @@
 #include <sys/wait.h>
 #include <sys/stat.h>
 #include <stdatomic.h>
+#include "llamafile/log.h"
 #include "llama.cpp/ggml-cuda.h"
 
 __static_yoink("llama.cpp/ggml.h");
@@ -160,20 +161,20 @@ static bool CreateTempPath(const char *path, char tmp[static PATH_MAX]) {
 static void LogCommand(char *args[]) {
     for (int i = 0; args[i]; ++i) {
         if (i) {
-            tinyprint(2, " ", NULL);
+            tinylog(" ", NULL);
         }
         // this quoting should be close enough to correct to be
         // copy/pastable on both unix and windows command terms
         bool need_quotes = !!strchr(args[i], ' ');
         if (need_quotes) {
-            tinyprint(2, "\"", NULL);
+            tinylog("\"", NULL);
         }
-        tinyprint(2, args[i], NULL);
+        tinylog(args[i], NULL);
         if (need_quotes) {
-            tinyprint(2, "\"", NULL);
+            tinylog("\"", NULL);
         }
     }
-    tinyprint(2, "\n", NULL);
+    tinylog("\n", NULL);
 }
 
 static bool Compile(const char *src,
@@ -196,7 +197,7 @@ static bool Compile(const char *src,
         }
     }
     if (ws) {
-        tinyprint(2, args[0], ": returned nonzero exit status\n", NULL);
+        tinylog(args[0], ": returned nonzero exit status\n", NULL);
         unlink(tmp);
         return false;
     }
@@ -318,13 +319,13 @@ static bool GetAmdOffloadArchFlag(char out[static 64]) {
         }
     }
     if (ws) {
-        tinyprint(2, "error: hipInfo returned non-zero exit status\n", NULL);
+        tinylog("error: hipInfo returned non-zero exit status\n", NULL);
         return false;
     }
 
     // Serialize value for --offload-arch=LIST flag.
     if (!archs) {
-        tinyprint(2, "warning: hipInfo output didn't list any graphics cards\n", NULL);
+        tinylog("warning: hipInfo output didn't list any graphics cards\n", NULL);
         return false;
     }
     bool gotsome = false;
@@ -396,7 +397,7 @@ static dontinline bool GetNvccArchFlag(const char *nvcc, char flag[static 32]) {
     // nvidia-smi in cuda 11.5+ can do this but (1) it takes longer to
     // run than compiling / running this script and (2) the nvidia-smi
     // command isn't available on Jetson devices.
-    tinyprint(2, "building nvidia compute capability detector...\n", NULL);
+    tinylog("building nvidia compute capability detector...\n", NULL);
     if (!Compile(src, tmp, exe, (char *[]){(char *)nvcc, "-o", tmp, src, 0})) {
         return false;
     }
@@ -434,14 +435,14 @@ static dontinline bool GetNvccArchFlag(const char *nvcc, char flag[static 32]) {
         }
     }
     if (ws) {
-        tinyprint(2, "error: compute capability detector returned nonzero exit status\n", NULL);
+        tinylog("error: compute capability detector returned nonzero exit status\n", NULL);
         return false;
     }
 
     // parse output of detector
     char *endptr;
     if (!*ibuf || !strtol(ibuf, &endptr, 10) || *endptr) {
-        tinyprint(2, "error: bad compute capability detector output\n", NULL);
+        tinylog("error: bad compute capability detector output\n", NULL);
         return false;
     }
 
@@ -528,7 +529,7 @@ static bool CompileNvidia(const char *nvcc, const char *dso, const char *src) {
     }
 
     // try building dso with host nvidia microarchitecture
-    tinyprint(2, "building ggml-cuda with nvcc -arch=native...\n", NULL);
+    tinylog("building ggml-cuda with nvcc -arch=native...\n", NULL);
     if (Compile(src, tmpdso, dso, (char *[]){
                 (char *)nvcc, "-arch=native", NVCC_FLAGS, "-o", tmpdso,
                 (char *)src, NVCC_LIBS, NULL})) {
@@ -540,7 +541,7 @@ static bool CompileNvidia(const char *nvcc, const char *dso, const char *src) {
     if (!GetNvccArchFlag(nvcc, archflag)) {
         return false;
     }
-    tinyprint(2, "building ggml-cuda with nvcc ", archflag, "...\n", NULL);
+    tinylog("building ggml-cuda with nvcc ", archflag, "...\n", NULL);
     if (Compile(src, tmpdso, dso, (char *[]){
                 (char *)nvcc, archflag, NVCC_FLAGS, "-o", tmpdso,
                 (char *)src, NVCC_LIBS, NULL})) {
@@ -560,7 +561,7 @@ static bool ExtractCudaDso(const char *dso, const char *name) {
     strlcat(zip, ".", sizeof(zip));
     strlcat(zip, GetDsoExtension(), sizeof(zip));
     if (!FileExists(zip)) {
-        tinyprint(2, "prebuilt binary ", zip, " not found\n", NULL);
+        tinylog("prebuilt binary ", zip, " not found\n", NULL);
         return false;
     }
 
@@ -579,16 +580,16 @@ static bool LinkCudaDso(const char *dso, const char *dir) {
 
     // runtime link dynamic shared object
     void *lib;
-    tinyprint(2, "dynamically linking ", dso, "\n", NULL);
+    tinylog("dynamically linking ", dso, "\n", NULL);
     lib = cosmo_dlopen(dso, RTLD_LAZY);
     if (dir) {
         chdir(cwd);
     }
     if (!lib) {
         char cc[PATH_MAX];
-        tinyprint(2, Dlerror(), ": failed to load library\n", NULL);
+        tinylog(Dlerror(), ": failed to load library\n", NULL);
         if ((IsLinux() || IsBsd()) && !commandv("cc", cc, PATH_MAX)) {
-            tinyprint(2, "you need to install cc for gpu support\n", NULL);
+            tinylog("you need to install cc for gpu support\n", NULL);
         }
         return false;
     }
@@ -618,7 +619,7 @@ static bool LinkCudaDso(const char *dso, const char *dir) {
     ok &= !!(ggml_cuda.reg_cuda_init = cosmo_dlsym(lib, "ggml_backend_reg_cuda_init"));
     ok &= !!(ggml_cuda.buffer_type = cosmo_dlsym(lib, "ggml_backend_cuda_buffer_type"));
     if (!ok) {
-        tinyprint(2, Dlerror(), ": not all symbols could be imported\n", NULL);
+        tinylog(Dlerror(), ": not all symbols could be imported\n", NULL);
         return false;
     }
 
@@ -643,7 +644,7 @@ static bool ImportCudaImpl(void) {
         default:
             return false;
     }
-    tinyprint(2, "initializing gpu module...\n", NULL);
+    tinylog("initializing gpu module...\n", NULL);
 
     // extract source code
     char src[PATH_MAX];
@@ -738,7 +739,7 @@ static bool ImportCudaImpl(void) {
                     }
                 }
             } else {
-                tinyprint(2, "note: won't compile AMD GPU support because $HIP_PATH/bin/clang++ is missing\n", NULL);
+                tinylog("note: won't compile AMD GPU support because $HIP_PATH/bin/clang++ is missing\n", NULL);
             }
 
             // Try extracting prebuilt tinyBLAS DSO from PKZIP.
@@ -818,7 +819,7 @@ static bool ImportCudaImpl(void) {
 
 static void ImportCuda(void) {
     if (!ggml_cuda.disabled && ImportCudaImpl()) {
-        tinyprint(2, "GPU support successfully linked and loaded\n", NULL);
+        tinylog("GPU support successfully linked and loaded\n", NULL);
         ggml_cuda.supported = true;
     }
 }

llamafile/extract.c

@@ -24,6 +24,7 @@
 #include <unistd.h>
 #include <sys/stat.h>
 #include "llamafile.h"
+#include "llamafile/log.h"
 
 /**
  * Returns true if `zip` was successfully copied to `to`.
@@ -35,7 +36,7 @@
 bool llamafile_extract(const char *zip, const char *to) {
     int fdin, fdout;
     char stage[PATH_MAX];
-    tinyprint(2, "extracting ", zip, " to ", to, "\n", NULL);
+    tinylog("extracting ", zip, " to ", to, "\n", NULL);
     strlcpy(stage, to, sizeof(stage));
     if (strlcat(stage, ".XXXXXX", sizeof(stage)) >= sizeof(stage)) {
         errno = ENAMETOOLONG;

llamafile/gpu.c

@@ -1,7 +1,7 @@
 // -*- mode:c;indent-tabs-mode:nil;c-basic-offset:4;coding:utf-8 -*-
 // vi: set et ft=c ts=4 sts=4 sw=4 fenc=utf-8 :vi
 //
-// Copyright 2023 Mozilla Foundation
+// Copyright 2024 Mozilla Foundation
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.

llamafile/launch_browser.c

@@ -22,12 +22,13 @@
 #include <stdio.h>
 #include <string.h>
 #include <sys/wait.h>
+#include "llamafile/log.h"
 
 static void report_failure(const char *url,
                            const char *cmd,
                            const char *reason) {
-    tinyprint(2, "failed to open ", url, " in a browser tab using ", cmd,
-              ": ", reason, "\n", NULL);
+    tinylog("failed to open ", url, " in a browser tab using ", cmd,
+            ": ", reason, "\n", NULL);
 }
 
 /**

llamafile/log.c

@@ -0,0 +1,43 @@
+// -*- mode:c;indent-tabs-mode:nil;c-basic-offset:4;coding:utf-8 -*-
+// vi: set et ft=c ts=4 sts=4 sw=4 fenc=utf-8 :vi
+//
+// Copyright 2024 Mozilla Foundation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "log.h"
+#include <unistd.h>
+#include <pthread.h>
+
+bool FLAG_log_disable;
+
+void (tinylog)(const char *s, ...) {
+  size_t n;
+  int c, cs;
+  va_list va;
+  char buf[512];
+  pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &cs);
+  va_start(va, s);
+  for (n = 0; s; s = va_arg(va, const char *)) {
+    while ((c = *s++)) {
+      buf[n++] = c;
+      if (n == sizeof(buf)) {
+        write(2, buf, n);
+        n = 0;
+      }
+    }
+  }
+  va_end(va);
+  write(2, buf, n);
+  pthread_setcancelstate(cs, 0);
+}

llamafile/log.h

@@ -0,0 +1,17 @@
+#ifndef LLAMAFILE_LOG_H_
+#define LLAMAFILE_LOG_H_
+#include <stdarg.h>
+#include <stdbool.h>
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+extern bool FLAG_log_disable;
+
+void tinylog(const char *, ...);
+#define tinylog(...) (void)(!FLAG_log_disable && (tinylog(__VA_ARGS__), 0))
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* LLAMAFILE_LOG_H_ */

llamafile/metal.c

@@ -130,7 +130,7 @@ static bool BuildMetal(const char *dso) {
 
     // compile dynamic shared object
     if (needs_rebuild) {
-        tinyprint(2, "building ggml-metal.dylib with xcode...\n", NULL);
+        tinylog("building ggml-metal.dylib with xcode...\n", NULL);
         int fd;
         char tmpdso[PATH_MAX];
         strlcpy(tmpdso, dso, sizeof(tmpdso));
@@ -162,7 +162,7 @@ static bool BuildMetal(const char *dso) {
         if (err) {
             perror("cc");
             if (err == ENOENT) {
-                tinyprint(2, "PLEASE RUN: xcode-select --install\n", NULL);
+                tinylog("PLEASE RUN: xcode-select --install\n", NULL);
             }
             return false;
         }
@@ -173,7 +173,7 @@ static bool BuildMetal(const char *dso) {
             }
         }
         if (ws) {
-            tinyprint(2, "compiler returned nonzero exit status\n", NULL);
+            tinylog("compiler returned nonzero exit status\n", NULL);
             return false;
         }
         if (rename(tmpdso, dso)) {
@@ -191,7 +191,7 @@ static bool LinkMetal(const char *dso) {
     void *lib;
     lib = cosmo_dlopen(dso, RTLD_LAZY);
     if (!lib) {
-        tinyprint(2, Dlerror(), ": failed to load library\n", NULL);
+        tinylog(Dlerror(), ": failed to load library\n", NULL);
         return false;
     }
 
@@ -209,7 +209,7 @@ static bool LinkMetal(const char *dso) {
     ok &= !!(ggml_metal.set_n_cb = cosmo_dlsym(lib, "ggml_metal_set_n_cb"));
     ok &= !!(ggml_metal.backend_init = cosmo_dlsym(lib, "ggml_backend_metal_init"));
     if (!ok) {
-        tinyprint(2, Dlerror(), ": not all symbols could be imported\n", NULL);
+        tinylog(Dlerror(), ": not all symbols could be imported\n", NULL);
         return false;
     }
 
@@ -253,7 +253,7 @@ static void ImportMetal(void) {
     if (ImportMetalImpl()) {
         ggml_metal.supported = true;
         ggml_metal.backend_init();
-        tinyprint(2, "Apple Metal GPU support successfully loaded\n", NULL);
+        tinylog("Apple Metal GPU support successfully loaded\n", NULL);
     }
 }